[Host] add sequence_pad (#5867) (#6088)

lite/kernels/host/CMakeLists.txt

@@ -47,6 +47,7 @@ add_kernel(sin_compute_host Host extra SRCS sin_compute.cc DEPS ${lite_kernel_de
 add_kernel(cos_compute_host Host extra SRCS cos_compute.cc DEPS ${lite_kernel_deps})
 add_kernel(crop_compute_host Host extra SRCS crop_compute.cc DEPS ${lite_kernel_deps} math_host)
 add_kernel(crop_tensor_compute_host Host extra SRCS crop_tensor_compute.cc DEPS ${lite_kernel_deps} math_host)
+add_kernel(sequence_pad_compute_host Host extra SRCS sequence_pad_compute.cc DEPS ${lite_kernel_deps} math_host)
 add_kernel(sequence_unpad_compute_host Host extra SRCS sequence_unpad_compute.cc DEPS ${lite_kernel_deps} math_host)
 add_kernel(flatten_contiguous_range_compute_host Host extra SRCS flatten_compute.cc DEPS ${lite_kernel_deps})
 add_kernel(shuffle_channel_compute_host Host extra SRCS shuffle_channel_compute.cc DEPS ${lite_kernel_deps} math_host)

lite/kernels/host/sequence_pad_compute.cc

@@ -0,0 +1,100 @@
+// Copyright (c) 2021 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.
+
+#include "lite/kernels/host/sequence_pad_compute.h"
+#include "lite/backends/host/math/sequence_padding.h"
+
+namespace paddle {
+namespace lite {
+namespace kernels {
+namespace host {
+
+template <class T>
+void SequencePadCompute<T>::Run() {
+  auto& param = this->template Param<param_t>();
+  auto& ctx = this->ctx_->template As<HostContext>();
+
+  auto* x = param.X;
+  auto* pad_value = param.PadValue;
+  auto* len_t = param.Length;
+  auto* out = param.Out;
+  CHECK(!x->lod().empty()) << "Input X should have lod data.";
+  int padded_length = param.padded_length;
+
+  lite::host::math::PaddingLoDTensorFunctor<lite::TargetType::kHost, T>()(
+      ctx,
+      *x,
+      out,
+      *pad_value,
+      padded_length,
+      0,
+      false,
+      lite::host::math::kBatchLengthWidth);
+
+  auto* len_data = len_t->template mutable_data<int64_t>();
+  auto x_lod = x->lod();
+  for (size_t i = 1; i < x_lod[0].size(); i++) {
+    len_data[i - 1] = x_lod[0][i] - x_lod[0][i - 1];
+  }
+}
+
+}  // namespace host
+}  // namespace kernels
+}  // namespace lite
+}  // namespace paddle
+
+REGISTER_LITE_KERNEL(sequence_pad,
+                     kHost,
+                     kFloat,
+                     kNCHW,
+                     paddle::lite::kernels::host::SequencePadCompute<float>,
+                     def)
+    .BindInput("X", {LiteType::GetTensorTy(TARGET(kHost), PRECISION(kFloat))})
+    .BindInput("PadValue",
+               {LiteType::GetTensorTy(TARGET(kHost), PRECISION(kFloat))})
+    .BindOutput("Out",
+                {LiteType::GetTensorTy(TARGET(kHost), PRECISION(kFloat))})
+    .BindOutput("Length",
+                {LiteType::GetTensorTy(TARGET(kHost), PRECISION(kInt64))})
+    .Finalize();
+
+REGISTER_LITE_KERNEL(sequence_pad,
+                     kHost,
+                     kFloat,
+                     kNCHW,
+                     paddle::lite::kernels::host::SequencePadCompute<int>,
+                     int32)
+    .BindInput("X", {LiteType::GetTensorTy(TARGET(kHost), PRECISION(kInt32))})
+    .BindInput("PadValue",
+               {LiteType::GetTensorTy(TARGET(kHost), PRECISION(kInt32))})
+    .BindOutput("Out",
+                {LiteType::GetTensorTy(TARGET(kHost), PRECISION(kInt32))})
+    .BindOutput("Length",
+                {LiteType::GetTensorTy(TARGET(kHost), PRECISION(kInt64))})
+    .Finalize();
+
+REGISTER_LITE_KERNEL(sequence_pad,
+                     kHost,
+                     kFloat,
+                     kNCHW,
+                     paddle::lite::kernels::host::SequencePadCompute<int64_t>,
+                     int64)
+    .BindInput("X", {LiteType::GetTensorTy(TARGET(kHost), PRECISION(kInt64))})
+    .BindInput("PadValue",
+               {LiteType::GetTensorTy(TARGET(kHost), PRECISION(kInt64))})
+    .BindOutput("Out",
+                {LiteType::GetTensorTy(TARGET(kHost), PRECISION(kInt64))})
+    .BindOutput("Length",
+                {LiteType::GetTensorTy(TARGET(kHost), PRECISION(kInt64))})
+    .Finalize();

lite/kernels/host/sequence_pad_compute.h

@@ -0,0 +1,37 @@
+// Copyright (c) 2021 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.
+
+#pragma once
+#include "lite/core/kernel.h"
+#include "lite/core/op_registry.h"
+
+namespace paddle {
+namespace lite {
+namespace kernels {
+namespace host {
+
+template <class T>
+class SequencePadCompute : public KernelLite<TARGET(kHost), PRECISION(kFloat)> {
+ public:
+  using param_t = operators::SequencePadParam;
+
+  void Run() override;
+
+  virtual ~SequencePadCompute() = default;
+};
+
+}  // namespace host
+}  // namespace kernels
+}  // namespace lite
+}  // namespace paddle

lite/tests/kernels/CMakeLists.txt

@@ -64,6 +64,7 @@ if((NOT LITE_WITH_FPGA AND NOT LITE_WITH_BM AND NOT LITE_WITH_MLU) AND (LITE_WIT
 if(LITE_BUILD_EXTRA)
     lite_cc_test(test_gru_unit SRCS gru_unit_test.cc DEPS ${test_kernel_deps})
     lite_cc_test(test_scatter_nd_add SRCS scatter_nd_add_test.cc DEPS ${test_kernel_deps})
+    lite_cc_test(test_sequence_pad SRCS sequence_pad_test.cc DEPS ${test_kernel_deps})
     #lite_cc_test(test_kernel_sequence_pool_compute SRCS sequence_pool_compute_test.cc DEPS ${test_kernel_deps})
     lite_cc_test(test_kernel_sequence_conv_compute SRCS sequence_conv_compute_test.cc DEPS ${test_kernel_deps})
     lite_cc_test(test_kernel_reduce_max_compute SRCS reduce_max_compute_test.cc DEPS ${test_kernel_deps})

lite/tests/kernels/sequence_pad_test.cc

@@ -0,0 +1,116 @@
+// Copyright (c) 2021 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.
+
+#include <gtest/gtest.h>
+#include <cstring>
+#include "lite/api/paddle_use_kernels.h"
+#include "lite/api/paddle_use_ops.h"
+#include "lite/core/arena/framework.h"
+#include "lite/tests/utils/fill_data.h"
+
+namespace paddle {
+namespace lite {
+
+template <class T>
+class SequencePadTester : public arena::TestCase {
+ protected:
+  std::string x_ = "x";
+  std::string pad_value_ = "pad_value";
+  std::string out_ = "out";
+  std::string length_ = "length";
+  DDim x_dims_{{9, 2, 3, 4}};
+  LoD x_lod_{{{0, 2, 5, 9}}};
+  T value_ = 0;
+  int padded_length_ = 4;
+
+ public:
+  SequencePadTester(const Place& place, const std::string& alias)
+      : TestCase(place, alias) {}
+
+  void RunBaseline(Scope* scope) override {
+    auto* out = scope->NewTensor(out_);
+    auto out_shape = x_dims_.Vectorize();
+    out_shape[0] = padded_length_;
+    out_shape.insert(out_shape.begin(),
+                     static_cast<int64_t>(x_lod_[0].size() - 1));
+    out->Resize(out_shape);
+    auto* out_data = out->template mutable_data<T>();
+    for (int64_t i = 0; i < out->numel(); i++) {
+      out_data[i] = value_;
+    }
+
+    int n = x_dims_.production() / x_dims_[0];
+    int out_step = padded_length_ * n;
+    auto* x = scope->FindTensor(x_);
+    auto* x_data = x->template data<T>();
+    for (size_t i = 1; i < x_lod_[0].size(); i++) {
+      int x_step = (x_lod_[0][i] - x_lod_[0][i - 1]) * n;
+      memcpy(out_data, x_data, sizeof(T) * x_step);
+      x_data += x_step;
+      out_data += out_step;
+    }
+
+    auto* length = scope->NewTensor(length_);
+    length->Resize({static_cast<int64_t>(x_lod_[0].size() - 1)});
+    int64_t* length_data = length->template mutable_data<int64_t>();
+    for (size_t i = 1; i < x_lod_[0].size(); i++) {
+      length_data[i - 1] = x_lod_[0][i] - x_lod_[0][i - 1];
+    }
+  }
+
+  void PrepareOpDesc(cpp::OpDesc* op_desc) {
+    op_desc->SetType("sequence_pad");
+    op_desc->SetInput("X", {x_});
+    op_desc->SetInput("PadValue", {pad_value_});
+    op_desc->SetOutput("Out", {out_});
+    op_desc->SetOutput("Length", {length_});
+    op_desc->SetAttr("padded_length", padded_length_);
+  }
+
+  void PrepareData() override {
+    std::vector<T> x_data(x_dims_.production());
+    fill_data_rand<T>(x_data.data(), -10, 10, x_dims_.production());
+    SetCommonTensor(x_, x_dims_, x_data.data(), x_lod_);
+
+    std::vector<T> pad_value_data{0};
+    SetCommonTensor(pad_value_, DDim{{1}}, pad_value_data.data());
+  }
+};
+
+template <class T>
+void TestSequencePad(const Place place,
+                     const float abs_error,
+                     const std::string alias) {
+  std::unique_ptr<arena::TestCase> tester(
+      new SequencePadTester<T>(place, alias));
+  arena::Arena arena(std::move(tester), place, abs_error);
+  arena.TestPrecision();
+}
+
+TEST(sequence_pad, precision) {
+  Place place;
+  float abs_error = 1e-5;
+#if defined(LITE_WITH_ARM) || defined(LITE_WITH_X86)
+  place = TARGET(kHost);
+#else
+  return;
+#endif
+
+  TestSequencePad<float>(place, abs_error, "def");
+  TestSequencePad<int>(place, abs_error, "int32");
+  TestSequencePad<int64_t>(place, abs_error, "int64");
+}
+
+}  // namespace lite
+}  // namespace paddle