Upgrade concat and unsqueeze, test=develop (#2378)

* update concat and unsqueeze, test=develop

lite/kernels/arm/concat_compute.cc

@@ -39,6 +39,11 @@ void ConcatCompute::Run() {
   std::vector<lite::Tensor*> inputs = param.x;
   auto* out = param.output;
   int axis = param.axis;
+  auto* axis_tensor = param.axis_tensor;
+  if (axis_tensor != nullptr) {
+    auto* axis_tensor_data = axis_tensor->data<int>();
+    axis = axis_tensor_data[0];
+  }
   out->mutable_data<float>();
 
   /// Sometimes direct copies will be faster, this maybe need deeply analysis.
@@ -83,5 +88,7 @@ void ConcatCompute::Run() {
 REGISTER_LITE_KERNEL(
     concat, kARM, kFloat, kNCHW, paddle::lite::kernels::arm::ConcatCompute, def)
     .BindInput("X", {LiteType::GetTensorTy(TARGET(kARM))})
+    .BindInput("AxisTensor",
+               {LiteType::GetTensorTy(TARGET(kARM), PRECISION(kInt32))})
     .BindOutput("Out", {LiteType::GetTensorTy(TARGET(kARM))})
     .Finalize();

lite/kernels/arm/unsqueeze_compute.cc

@@ -55,6 +55,10 @@ REGISTER_LITE_KERNEL(unsqueeze,
                      paddle::lite::kernels::host::UnsqueezeCompute,
                      def)
     .BindInput("X", {LiteType::GetTensorTy(TARGET(kARM))})
+    .BindInput("AxesTensor",
+               {LiteType::GetTensorTy(TARGET(kARM), PRECISION(kInt32))})
+    .BindInput("AxesTensorList",
+               {LiteType::GetTensorTy(TARGET(kARM), PRECISION(kInt32))})
     .BindOutput("Out", {LiteType::GetTensorTy(TARGET(kARM))})
     .Finalize();
 
@@ -65,6 +69,10 @@ REGISTER_LITE_KERNEL(unsqueeze2,
                      paddle::lite::kernels::host::Unsqueeze2Compute,
                      def)
     .BindInput("X", {LiteType::GetTensorTy(TARGET(kARM))})
+    .BindInput("AxesTensor",
+               {LiteType::GetTensorTy(TARGET(kARM), PRECISION(kInt32))})
+    .BindInput("AxesTensorList",
+               {LiteType::GetTensorTy(TARGET(kARM), PRECISION(kInt32))})
     .BindOutput("Out", {LiteType::GetTensorTy(TARGET(kARM))})
     .BindOutput("XShape", {LiteType::GetTensorTy(TARGET(kARM))})
     .Finalize();

lite/kernels/cuda/concat_compute.cu

@@ -51,6 +51,11 @@ void ConcatCompute<Dtype>::Run() {
   Tensor* output = param.output;
   auto* output_data = output->mutable_data<Dtype>(TARGET(kCUDA));
   int axis = param.axis;
+  auto* axis_tensor = param.axis_tensor;
+  if (axis_tensor != nullptr) {
+    auto* axis_tensor_data = axis_tensor->data<int>();
+    axis = axis_tensor_data[0];
+  }
   int inner_size = 1;
   int outer_size = 1;
   auto input_dims = input[0]->dims();
@@ -97,5 +102,7 @@ REGISTER_LITE_KERNEL(concat,
                      paddle::lite::kernels::cuda::ConcatCompute<float>,
                      def)
     .BindInput("X", {LiteType::GetTensorTy(TARGET(kCUDA))})
+    .BindInput("AxisTensor",
+               {LiteType::GetTensorTy(TARGET(kCUDA), PRECISION(kInt32))})
     .BindOutput("Out", {LiteType::GetTensorTy(TARGET(kCUDA))})
     .Finalize();

lite/kernels/x86/concat_compute.cc

@@ -21,5 +21,7 @@ REGISTER_LITE_KERNEL(concat,
                      paddle::lite::kernels::x86::ConcatCompute<float>,
                      def)
     .BindInput("X", {LiteType::GetTensorTy(TARGET(kX86))})
+    .BindInput("AxisTensor",
+               {LiteType::GetTensorTy(TARGET(kX86), PRECISION(kInt32))})
     .BindOutput("Out", {LiteType::GetTensorTy(TARGET(kX86))})
     .Finalize();

lite/kernels/x86/concat_compute.h

@@ -40,6 +40,11 @@ class ConcatCompute : public KernelLite<TARGET(kX86), PRECISION(kFloat)> {
   void Run() override {
     auto& param = *param_.get_mutable<param_t>();
     int64_t axis = static_cast<int64_t>(param.axis);
+    auto* axis_tensor = param.axis_tensor;
+    if (axis_tensor != nullptr) {
+      auto* axis_tensor_data = axis_tensor->data<int>();
+      axis = static_cast<int64_t>(axis_tensor_data[0]);
+    }
     auto x_dims = param.x[0]->dims();
     auto out = param.output;
     if (param.x.size() == 1) {

lite/operators/concat_op.cc

@@ -31,14 +31,25 @@ bool ConcatOpLite::InferShape() const {
   for (auto p : param_.x) {
     input_dims.push_back(p->dims());
   }
-  size_t axis = static_cast<size_t>(param_.axis);
   const size_t n = input_dims.size();
   CHECK_GT_OR_FALSE(n, 0);
+
+  int axis = 0;
+  if (param_.axis_tensor == nullptr) {
+    axis = param_.axis;
+  } else {
+    auto *axis_tensor_val = param_.axis_tensor->data<int>();
+    axis = axis_tensor_val[0];
+  }
+  if (axis < 0) {
+    axis += input_dims[0].size();
+  }
+
   auto &out_dims = input_dims[0];
   size_t in_zero_dims_size = out_dims.size();
   for (size_t i = 1; i < n; i++) {
     for (size_t j = 0; j < in_zero_dims_size; j++) {
-      if (j == axis) {
+      if (j == static_cast<size_t>(axis)) {
         out_dims[axis] += input_dims[i][j];
       } else {
         CHECK_EQ_OR_FALSE(out_dims[j], input_dims[i][j]);
@@ -68,6 +79,17 @@ bool ConcatOpLite::AttachImpl(const cpp::OpDesc &op_desc, lite::Scope *scope) {
   param_.output = scope->FindVar(out)->GetMutable<lite::Tensor>();
   param_.axis = op_desc.GetAttr<int>("axis");
 
+  std::vector<std::string> input_arg_names = op_desc.InputArgumentNames();
+  if (std::find(input_arg_names.begin(), input_arg_names.end(), "AxisTensor") !=
+      input_arg_names.end()) {
+    auto arguments = op_desc.Input("AxisTensor");
+    if (arguments.size() > 0) {
+      auto var = scope->FindVar(arguments.front());
+      if (var != nullptr) {
+        param_.axis_tensor = var->GetMutable<lite::Tensor>();
+      }
+    }
+  }
   return true;
 }
 

lite/operators/op_params.h

@@ -207,6 +207,7 @@ struct ConcatParam {
   std::vector<lite::Tensor*> x{};
   lite::Tensor* output{};
   int axis{0};
+  lite::Tensor* axis_tensor{};
 };
 
 /// ----------------------- activation operators ----------------------
@@ -854,6 +855,8 @@ struct UnsqueezeParam {
   lite::Tensor* Out{};
   lite::Tensor* XShape{};
   std::vector<int> axes{};
+  const lite::Tensor* axes_tensor{};
+  std::vector<lite::Tensor>* axes_tensor_vct{};
 };
 
 /// ----------------------- expand operators ----------------------

lite/operators/unsqueeze_op.cc

@@ -63,9 +63,30 @@ bool UnsqueezeOp::CheckShape() const {
 }
 
 bool UnsqueezeOp::InferShape() const {
-  std::vector<int> unsqueeze_dims = param_.axes;
+  std::vector<int> final_axes;
+  auto axes = param_.axes;
+  auto *axes_tensor = param_.axes_tensor;
+  std::vector<lite::Tensor> axes_tensor_vct;
+  if (param_.axes_tensor_vct) {
+    axes_tensor_vct = *(param_.axes_tensor_vct);
+  }
+
+  if (!axes.empty()) {
+    final_axes = axes;
+  } else if (axes_tensor != nullptr) {
+    auto *axes_tensor_data = axes_tensor->data<int>();
+    final_axes = std::vector<int>(axes_tensor_data,
+                                  axes_tensor_data + axes_tensor->numel());
+  } else if (!axes_tensor_vct.empty()) {
+    for (int i = 0; i < axes_tensor_vct.size(); i++) {
+      final_axes.push_back(axes_tensor_vct[i].data<int>()[0]);
+    }
+  } else {
+    LOG(FATAL) << "Input axis error";
+  }
+
   DDim in_dims = param_.X->dims();
-  DDim out_dim = GetOutputShape(unsqueeze_dims, in_dims);
+  DDim out_dim = GetOutputShape(final_axes, in_dims);
   param_.Out->Resize(out_dim);
   return true;
 }
@@ -81,6 +102,29 @@ bool UnsqueezeOp::AttachImpl(const cpp::OpDesc &opdesc, lite::Scope *scope) {
   if (opdesc.HasAttr("axes")) {
     param_.axes = opdesc.GetAttr<std::vector<int>>("axes");
   }
+
+  if (opdesc.HasInput("AxesTensor") && opdesc.Input("AxesTensor").size() > 0) {
+    auto var = scope->FindVar(opdesc.Input("AxesTensor").front());
+    if (var != nullptr) {
+      param_.axes_tensor = var->GetMutable<lite::Tensor>();
+      VLOG(5) << "load AxesTensor";
+    }
+  }
+
+  if (opdesc.HasInput("AxesTensorList") &&
+      opdesc.Input("AxesTensorList").size() > 0) {
+    auto args = opdesc.Input("AxesTensorList");
+    /*
+    for (auto arg : args) {
+      auto *var = scope->FindVar(arg);
+      if (var != nullptr) {
+        param_.axes_tensor_vct.push_back(var->GetMutable<lite::Tensor>());
+      }
+    }
+    */
+    auto *var = scope->FindVar(args.front());
+    param_.axes_tensor_vct = var->GetMutable<std::vector<lite::Tensor>>();
+  }
   CHECK(param_.X) << "Input(X) of UnsqueezeOp should not be null.";
   CHECK(param_.Out) << "Output(Out) of UnsqueezeOp should not be null.";
   return true;

lite/tests/kernels/CMakeLists.txt

@@ -22,6 +22,7 @@ if((NOT LITE_WITH_OPENCL AND NOT LITE_WITH_FPGA AND NOT LITE_WITH_XPU) AND (LITE
     #lite_cc_test(test_kernel_increment_compute SRCS increment_compute_test.cc DEPS arena_framework ${x86_kernels} ${arm_kernels} ${lite_ops} ${host_kernels})
     #lite_cc_test(test_kernel_write_to_array_compute SRCS write_to_array_compute_test.cc DEPS arena_framework ${x86_kernels} ${arm_kernels} ${lite_ops} ${host_kernels})
     #lite_cc_test(test_kernel_read_from_array_compute SRCS read_from_array_compute_test.cc DEPS arena_framework ${x86_kernels} ${arm_kernels} ${lite_ops} ${host_kernels})
+    lite_cc_test(test_concat_compute SRCS concat_compute_test.cc DEPS arena_framework ${x86_kernels} ${arm_kernels} ${lite_ops} ${host_kernels})
 
 if(LITE_BUILD_EXTRA)
     lite_cc_test(test_gru_unit SRCS gru_unit_test.cc DEPS arena_framework ${x86_kernels} ${arm_kernels} ${lite_ops} ${host_kernels})

lite/tests/kernels/concat_compute_test.cc

+// Copyright (c) 2019 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 "lite/api/paddle_use_kernels.h"
+#include "lite/api/paddle_use_ops.h"
+#include "lite/core/arena/framework.h"
+
+namespace paddle {
+namespace lite {
+
+DDim infer_shape(const std::vector<const Tensor*>& inputs, int in_axis) {
+  std::vector<DDim> input_dims;
+  for (auto* tensor : inputs) {
+    input_dims.push_back(tensor->dims());
+  }
+  size_t axis = static_cast<size_t>(in_axis);
+
+  DDim out_dims = input_dims[0];
+  for (size_t i = 1; i < input_dims.size(); i++) {
+    for (size_t j = 0; j < input_dims[0].size(); j++) {
+      if (j == axis) {
+        out_dims[axis] += input_dims[i][j];
+      } else {
+        if (out_dims[j] != input_dims[i][j]) {
+          LOG(FATAL) << "infer shape error.";
+        }
+      }
+    }
+  }
+  if (out_dims[axis] < 0) {
+    out_dims[axis] = -1;
+  }
+
+  return out_dims;
+}
+
+class ConcateComputeTester : public arena::TestCase {
+ protected:
+  // common attributes for this op.
+  std::vector<std::string> x_vct_{};
+  std::string out_ = "out";
+  std::string axis_tensor_ = "axis_tensor";
+  int axis_ = 0;
+  bool is_use_axis_tensor_ = false;
+
+  int x_num_ = 3;
+  DDim x_dims_{{2, 3, 4, 5}};
+
+ public:
+  ConcateComputeTester(const Place& place,
+                       const std::string& alias,
+                       int axis,
+                       bool is_use_axis_tensor)
+      : TestCase(place, alias) {
+    axis_ = axis;
+    is_use_axis_tensor_ = is_use_axis_tensor;
+  }
+
+  void RunBaseline(Scope* scope) override {
+    std::vector<const Tensor*> x_vct;
+    for (std::string& name : x_vct_) {
+      x_vct.push_back(scope->FindTensor(name));
+    }
+
+    auto* out = scope->NewTensor(out_);
+    DDim output_dims = infer_shape(x_vct, axis_);
+    out->Resize(output_dims);
+    auto* output_data = out->mutable_data<float>();
+
+    int num = x_vct.size();
+    int rows = 1;
+    auto dim_0 = x_vct[0]->dims();
+    for (int i = 0; i < axis_; ++i) {
+      rows *= dim_0[i];
+    }
+    int out_rows = rows, out_cols = 0;
+
+    std::vector<int> input_cols(x_vct.size());
+    for (int i = 0; i < num; ++i) {
+      int input_i_numel = x_vct[i]->dims().size() == 0 ? 0 : 1;
+      for (int didx = 0; didx < x_vct[i]->dims().size(); ++didx) {
+        input_i_numel *= x_vct[i]->dims()[didx];
+      }
+      int t_cols = input_i_numel / rows;
+      out_cols += t_cols;
+      input_cols[i] = t_cols;
+    }
+
+    // computation
+    int col_idx = 0;
+    for (int j = 0; j < num; ++j) {
+      int col_len = input_cols[j];
+      auto input_data = x_vct[j]->data<float>();
+      for (int k = 0; k < out_rows; ++k) {
+        memcpy(output_data + k * out_cols + col_idx,
+               input_data + k * col_len,
+               sizeof(float) * col_len);
+      }
+      col_idx += col_len;
+    }
+  }
+
+  void PrepareOpDesc(cpp::OpDesc* op_desc) {
+    op_desc->SetType("concat");
+    op_desc->SetInput("X", x_vct_);
+    op_desc->SetAttr("axis", axis_);
+    if (is_use_axis_tensor_) {
+      op_desc->SetInput("AxisTensor", {axis_tensor_});
+    }
+    op_desc->SetOutput("Out", {out_});
+  }
+
+  void PrepareData() override {
+    for (int n = 0; n < x_num_; n++) {
+      std::vector<float> x_data(x_dims_.production());
+      for (int i = 0; i < x_dims_.production(); i++) {
+        x_data[i] = static_cast<float>(i + n);
+      }
+      const std::string x_name = "x_tensor_" + std::to_string(n);
+      x_vct_.push_back(x_name);
+      SetCommonTensor(x_name, x_dims_, x_data.data());
+    }
+
+    if (is_use_axis_tensor_) {
+      SetCommonTensor(axis_tensor_, DDim({1}), &axis_);
+      LOG(INFO) << "set axis tensor";
+    }
+  }
+};
+
+TEST(Concat, precision) {
+  LOG(INFO) << "test concat op, kARM";
+#ifdef LITE_WITH_ARM
+  Place place(TARGET(kARM));
+  for (int axis : {1, 2}) {
+    for (bool is_use_axis_tensor : {false, true}) {
+      LOG(INFO) << "axis:" << axis
+                << ", is_use_axis_tensor:" << is_use_axis_tensor;
+      std::unique_ptr<arena::TestCase> tester(
+          new ConcateComputeTester(place, "def", axis, is_use_axis_tensor));
+      arena::Arena arena(std::move(tester), place, 2e-5);
+      arena.TestPrecision();
+    }
+  }
+#endif
+
+#ifdef LITE_WITH_X86
+  Place place(TARGET(kX86));
+  LOG(INFO) << "test concate op, x86";
+  for (int axis : {1, 2}) {
+    for (bool is_use_axis_tensor : {false, true}) {
+      LOG(INFO) << "axis:" << axis
+                << ", is_use_axis_tensor:" << is_use_axis_tensor;
+      std::unique_ptr<arena::TestCase> tester(
+          new ConcateComputeTester(place, "def", axis, is_use_axis_tensor));
+      arena::Arena arena(std::move(tester), place, 2e-5);
+      arena.TestPrecision();
+    }
+  }
+
+#endif
+}
+
+}  // namespace lite
+}  // namespace paddle

lite/tests/kernels/unsqueeze_compute_test.cc

@@ -13,10 +13,10 @@
 // limitations under the License.
 
 #include <gtest/gtest.h>
+#include <string>
 #include "lite/api/paddle_use_kernels.h"
 #include "lite/api/paddle_use_ops.h"
 #include "lite/core/arena/framework.h"
-
 namespace paddle {
 namespace lite {
 
@@ -25,15 +25,24 @@ class UnsqueezeComputeTester : public arena::TestCase {
   // common attributes for this op.
   std::string x_ = "X";
   std::string out_ = "Out";
+  std::string axes_tensor_ = "AxesTensor";
+  std::vector<std::string> axes_tensor_list_;
   std::vector<int> axes_;
   DDim dims_;
+  // input_axes_flag_: 1 for axes, 2 for axes_tensor, 3 for axes_tensor_list
+  int input_axes_flag_ = 1;
 
  public:
   UnsqueezeComputeTester(const Place& place,
                          const std::string& alias,
                          const std::vector<int>& axes,
-                         DDim dims)
-      : TestCase(place, alias), axes_(axes), dims_(dims) {}
+                         DDim dims,
+                         int input_axes_flag)
+      : TestCase(place, alias), dims_(dims), input_axes_flag_(input_axes_flag) {
+    for (int v : axes) {
+      axes_.push_back(v);
+    }
+  }
 
   void RunBaseline(Scope* scope) override {
     const auto* input = scope->FindTensor(x_);
@@ -86,7 +95,15 @@ class UnsqueezeComputeTester : public arena::TestCase {
     op_desc->SetType("unsqueeze");
     op_desc->SetInput("X", {x_});
     op_desc->SetOutput("Out", {out_});
-    op_desc->SetAttr("axes", axes_);
+    if (input_axes_flag_ == 1) {
+      op_desc->SetAttr("axes", axes_);
+    } else if (input_axes_flag_ == 2) {
+      op_desc->SetInput("AxesTensor", {axes_tensor_});
+    } else if (input_axes_flag_ == 3) {
+      op_desc->SetInput("AxesTensorList", axes_tensor_list_);
+    } else {
+      LOG(FATAL) << "input input_axes_flag_ error. " << input_axes_flag_;
+    }
   }
 
   void PrepareData() override {
@@ -95,6 +112,23 @@ class UnsqueezeComputeTester : public arena::TestCase {
       in_data[i] = i;
     }
     SetCommonTensor(x_, dims_, in_data.data());
+
+    if (input_axes_flag_ == 2) {
+      DDim axes_tensor_dim{{static_cast<int>(axes_.size())}};
+      std::vector<int> axes_tensor_data(axes_.size());
+      for (int i = 0; i < axes_tensor_dim.production(); i++) {
+        axes_tensor_data[i] = axes_[i];
+      }
+      SetCommonTensor(axes_tensor_, axes_tensor_dim, axes_tensor_data.data());
+    } else if (input_axes_flag_ == 3) {
+      std::string name = "axes_tensor_";
+      for (size_t i = 0; i < axes_.size(); i++) {
+        name = name + std::to_string(i);
+        axes_tensor_list_.push_back(name);
+        std::vector<int> in_data = {axes_[i]};
+        SetCommonTensor(name, DDim({1}), in_data.data());
+      }
+    }
   }
 };
 
@@ -189,17 +223,22 @@ class Unsqueeze2ComputeTester : public arena::TestCase {
 };
 
 void test_unsqueeze(Place place) {
-  for (std::vector<int> axes : {std::vector<int>({}),
+  for (std::vector<int> axes : {std::vector<int>({1}),
                                 std::vector<int>({0, 2}),
                                 std::vector<int>({0, -2})}) {
     for (int N : {1}) {
       for (int C : {3}) {
         for (int H : {1}) {
           for (int W : {5}) {
-            std::unique_ptr<arena::TestCase> tester(new UnsqueezeComputeTester(
-                place, "def", axes, DDim({N, C, H, W})));
-            arena::Arena arena(std::move(tester), place, 2e-5);
-            arena.TestPrecision();
+            for (int input_axes_flag : {1, 2}) {
+              LOG(INFO) << N << " " << C << " " << H << " " << W << " "
+                        << input_axes_flag;
+              std::unique_ptr<arena::TestCase> tester(
+                  new UnsqueezeComputeTester(
+                      place, "def", axes, DDim({N, C, H, W}), input_axes_flag));
+              arena::Arena arena(std::move(tester), place, 2e-5);
+              arena.TestPrecision();
+            }
           }
         }
       }
@@ -208,7 +247,7 @@ void test_unsqueeze(Place place) {
 }
 
 void test_unsqueeze2(Place place) {
-  for (std::vector<int> axes : {std::vector<int>({}),
+  for (std::vector<int> axes : {std::vector<int>({0}),
                                 std::vector<int>({0, 2}),
                                 std::vector<int>({0, -2})}) {
     for (int N : {1}) {