[HOST] add tril_triu; fix expand_as (#5507) (#5720)

Co-authored-by: zhupengyang <[email protected]>

Author: BillDior

lite/kernels/host/CMakeLists.txt

@@ -63,6 +63,7 @@ add_kernel(topk_v2_compute_host Host extra SRCS topk_v2_compute.cc DEPS ${lite_k
 add_kernel(fill_any_like_compute_host Host extra SRCS fill_any_like_compute.cc DEPS ${lite_kernel_deps})
 add_kernel(meshgrid_compute_host Host extra SRCS meshgrid_compute.cc DEPS ${lite_kernel_deps})
 add_kernel(linspace_compute_host Host extra SRCS linspace_compute.cc DEPS ${lite_kernel_deps})
+add_kernel(tril_triu_compute_host Host extra SRCS tril_triu_compute.cc DEPS ${lite_kernel_deps})
 
 if(LITE_BUILD_EXTRA AND LITE_WITH_x86)
   lite_cc_test(test_where_index_compute_host SRCS where_index_compute.cc DEPS where_index_compute_host)

lite/kernels/host/expand_as_compute.cc

@@ -30,7 +30,6 @@ void ExpandAsCompute<T, PType>::Run() {
   const T* src = x->template data<T>();
   T* dst = out->template mutable_data<T>();
 
-  // int dims = expand_times.size();
   for (int i = 0; i < target->dims().size(); ++i) {
     int times = target->dims()[i] / x->dims()[i];
     expand_times.push_back(times);
@@ -75,12 +74,29 @@ REGISTER_LITE_KERNEL(expand_as, kHost, kFloat, kAny, expand_as_float, def)
                {LiteType::GetTensorTy(TARGET(kHost),
                                       PRECISION(kFloat),
                                       DATALAYOUT(kAny))})
-    .BindInput("Target",
+    .BindInput("target_tensor",
                {LiteType::GetTensorTy(TARGET(kHost),
-                                      PRECISION(kFloat),
+                                      PRECISION(kAny),
                                       DATALAYOUT(kAny))})
     .BindOutput("Out",
                 {LiteType::GetTensorTy(TARGET(kHost),
                                        PRECISION(kFloat),
                                        DATALAYOUT(kAny))})
     .Finalize();
+
+using expand_as_int64 =
+    paddle::lite::kernels::host::ExpandAsCompute<int64_t, PRECISION(kFloat)>;
+REGISTER_LITE_KERNEL(expand_as, kHost, kFloat, kAny, expand_as_int64, int64)
+    .BindInput("X",
+               {LiteType::GetTensorTy(TARGET(kHost),
+                                      PRECISION(kInt64),
+                                      DATALAYOUT(kAny))})
+    .BindInput("target_tensor",
+               {LiteType::GetTensorTy(TARGET(kHost),
+                                      PRECISION(kAny),
+                                      DATALAYOUT(kAny))})
+    .BindOutput("Out",
+                {LiteType::GetTensorTy(TARGET(kHost),
+                                       PRECISION(kInt64),
+                                       DATALAYOUT(kAny))})
+    .Finalize();

lite/kernels/host/tril_triu_compute.cc

@@ -0,0 +1,72 @@
+// 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/tril_triu_compute.h"
+
+namespace paddle {
+namespace lite {
+namespace kernels {
+namespace host {
+
+template <class T>
+void TrilTriu(const T* in,
+              const int64_t diagonal,
+              const bool lower,
+              const int64_t h,
+              const int64_t w,
+              T* out) {
+  int64_t size = h * w;
+  for (int64_t idx = 0; idx < size; idx++) {
+    const int64_t row = idx / w;
+    const int64_t col = idx % w;
+    const bool mask = lower ? (col - row > diagonal) : (col - row < diagonal);
+    out[idx] = mask ? 0 : in[idx];
+  }
+  return;
+}
+
+template <class T>
+void TrilTriuCompute<T>::Run() {
+  auto& param = this->template Param<param_t>();
+  const lite::Tensor* x = param.x;
+  lite::Tensor* out = param.out;
+  int64_t diagonal = param.diagonal;
+  bool lower = param.lower;
+
+  const T* x_data = x->template data<T>();
+  T* out_data = out->template mutable_data<T>();
+  auto x_dims = x->dims();
+  int64_t h = x_dims[x_dims.size() - 2];
+  int64_t w = x_dims[x_dims.size() - 1];
+  int64_t n = x_dims.production() / h / w;
+
+  for (int64_t i = 0; i < n; i++) {
+    TrilTriu(x_data, diagonal, lower, h, w, out_data);
+    x_data += h * w;
+    out_data += h * w;
+  }
+  return;
+}
+
+}  // namespace host
+}  // namespace kernels
+}  // namespace lite
+}  // namespace paddle
+
+using TrilTriuFloat32 = paddle::lite::kernels::host::TrilTriuCompute<float>;
+REGISTER_LITE_KERNEL(tril_triu, kHost, kAny, kNCHW, TrilTriuFloat32, float32)
+    .BindInput("X", {LiteType::GetTensorTy(TARGET(kHost), PRECISION(kFloat))})
+    .BindOutput("Out",
+                {LiteType::GetTensorTy(TARGET(kHost), PRECISION(kFloat))})
+    .Finalize();

lite/kernels/host/tril_triu_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 TrilTriuCompute : public KernelLite<TARGET(kHost), PRECISION(kAny)> {
+ public:
+  using param_t = operators::TrilTriuParam;
+
+  void Run() override;
+
+  virtual ~TrilTriuCompute() = default;
+};
+
+}  // namespace host
+}  // namespace kernels
+}  // namespace lite
+}  // namespace paddle

lite/operators/CMakeLists.txt

@@ -143,9 +143,9 @@ add_operator(select_input_op extra SRCS select_input_op.cc DEPS ${op_DEPS})
 add_operator(tensor_array_to_tensor_op extra SRCS tensor_array_to_tensor_op.cc DEPS ${op_DEPS})
 add_operator(expand_v2_op_lite extra SRCS expand_v2_op.cc DEPS ${op_DEPS})
 add_operator(tile_op extra SRCS tile_op.cc DEPS ${op_DEPS})
-add_operator(sum_op extra SRCS sum_op.cc DEPS ${op_DEPS})
 add_operator(meshgrid_op_lite extra SRCS meshgrid_op.cc DEPS ${op_DEPS})
 add_operator(linspace_op extra SRCS linspace_op.cc DEPS ${op_DEPS})
+add_operator(tril_triu_op extra SRCS tril_triu_op.cc DEPS ${op_DEPS})
 
 # for OCR specific
 add_operator(while_op extra SRCS while_op.cc DEPS ${op_DEPS})

lite/operators/expand_as_op.cc

@@ -48,7 +48,7 @@ bool ExpandAsOpLite::AttachImpl(const cpp::OpDesc& opdesc, lite::Scope* scope) {
   auto Out_name = opdesc.Output("Out").front();
   param_.X = GetVar<lite::Tensor>(scope, X_name);
   param_.Out = GetMutableVar<lite::Tensor>(scope, Out_name);
-  auto Target_name = opdesc.Input("Target").front();
+  auto Target_name = opdesc.Input("target_tensor").front();
   param_.Target = GetVar<lite::Tensor>(scope, Target_name);
   return true;
 }

lite/operators/op_params.h

@@ -646,6 +646,14 @@ struct TransposeParam : ParamBase {
   }
 };
 
+struct TrilTriuParam : ParamBase {
+  const lite::Tensor* x{nullptr};
+  lite::Tensor* out{nullptr};
+
+  int diagonal{0};
+  bool lower{true};
+};
+
 /// ----------------------- element wise operators ----------------------
 struct ElementwiseParam : ParamBase {
   const lite::Tensor* X{};

lite/operators/tril_triu_op.cc

@@ -0,0 +1,48 @@
+// 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/operators/tril_triu_op.h"
+#include "lite/core/op_registry.h"
+
+namespace paddle {
+namespace lite {
+namespace operators {
+
+bool TrilTriuOp::CheckShape() const {
+  CHECK(param_.x);
+  CHECK(param_.out);
+  return true;
+}
+
+bool TrilTriuOp::InferShapeImpl() const {
+  CHECK_GE(param_.x->dims().size(), 2UL);
+  param_.out->Resize(param_.x->dims());
+  param_.out->set_lod(param_.x->lod());
+  return true;
+}
+
+bool TrilTriuOp::AttachImpl(const cpp::OpDesc &op_desc, lite::Scope *scope) {
+  param_.x = scope->FindTensor(op_desc.Input("X").front());
+  param_.out = scope->FindMutableTensor(op_desc.Output("Out").front());
+
+  param_.diagonal = op_desc.GetAttr<int>("diagonal");
+  param_.lower = op_desc.GetAttr<bool>("lower");
+  return true;
+}
+
+}  // namespace operators
+}  // namespace lite
+}  // namespace paddle
+
+REGISTER_LITE_OP(tril_triu, paddle::lite::operators::TrilTriuOp);

lite/operators/tril_triu_op.h

@@ -0,0 +1,45 @@
+// 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 <string>
+#include "lite/core/op_lite.h"
+#include "lite/core/scope.h"
+#include "lite/utils/all.h"
+
+namespace paddle {
+namespace lite {
+namespace operators {
+
+class TrilTriuOp : public OpLite {
+ public:
+  TrilTriuOp() {}
+  explicit TrilTriuOp(const std::string &op_type) : OpLite(op_type) {}
+
+  bool CheckShape() const override;
+
+  bool InferShapeImpl() const override;
+
+  bool AttachImpl(const cpp::OpDesc &opdesc, lite::Scope *scope) override;
+
+  void AttachKernel(KernelBase *kernel) override { kernel->SetParam(param_); }
+  std::string DebugString() const override { return "tril_triu"; }
+
+ private:
+  mutable TrilTriuParam param_;
+};
+
+}  // namespace operators
+}  // namespace lite
+}  // namespace paddle

lite/tests/kernels/CMakeLists.txt

@@ -99,6 +99,7 @@ if(LITE_BUILD_EXTRA)
     lite_cc_test(test_kernel_sequence_expand_as_compute SRCS sequence_expand_as_compute_test.cc DEPS ${test_kernel_deps})
     lite_cc_test(test_kernel_sin_compute SRCS sin_compute_test.cc DEPS arena_framework ${test_kernel_deps})
     lite_cc_test(test_kernel_cos_compute SRCS cos_compute_test.cc DEPS arena_framework ${test_kernel_deps})
+    lite_cc_test(test_kernel_tril_triu_compute SRCS tril_triu_compute_test.cc DEPS arena_framework ${test_kernel_deps})
     lite_cc_test(test_kernel_pad3d_compute SRCS  pad3d_compute_test.cc DEPS arena_framework ${test_kernel_deps})
     lite_cc_test(test_kernel_select_input_compute SRCS select_input_compute_test.cc DEPS arena_framework ${test_kernel_deps})
     # lite_cc_test(test_kernel_tensor_array_to_tensor_compute SRCS tensor_array_to_tensor_compute_test.cc DEPS arena_framework ${test_kernel_deps})