[ARM] add instance norm op and ut, test=develop (#2578)

lite/backends/arm/math/interpolate.cc

@@ -477,17 +477,23 @@ void nearest_interp(const float* src,
   float scale_h_new = (with_align)
                           ? (static_cast<float>(h_in - 1) / (h_out - 1))
                           : (static_cast<float>(h_in) / (h_out));
-
-#pragma omp parallel for collapse(2) schedule(static)
-  for (int h = 0; h < h_out; ++h) {
-    for (int w = 0; w < w_out; ++w) {
-      int near_x = (with_align) ? static_cast<int>(scale_w_new * w + 0.5)
-                                : static_cast<int>(scale_w_new * w);
-      int near_y = (with_align) ? static_cast<int>(scale_h_new * h + 0.5)
-                                : static_cast<int>(scale_h_new * h);
-      near_x = near_x < 0 ? 0 : near_x;
-      near_y = near_y < 0 ? 0 : near_y;
-      dst[h * w_out + w] = src[near_y * w_in + near_x];
+  if (with_align) {
+    for (int h = 0; h < h_out; ++h) {
+      float* dst_p = dst + h * w_out;
+      int near_y = static_cast<int>(scale_h_new * h + 0.5);
+      for (int w = 0; w < w_out; ++w) {
+        int near_x = static_cast<int>(scale_w_new * w + 0.5);
+        *dst_p++ = src[near_y * w_in + near_x];
+      }
+    }
+  } else {
+    for (int h = 0; h < h_out; ++h) {
+      float* dst_p = dst + h * w_out;
+      int near_y = static_cast<int>(scale_h_new * h);
+      for (int w = 0; w < w_out; ++w) {
+        int near_x = static_cast<int>(scale_w_new * w);
+        *dst_p++ = src[near_y * w_in + near_x];
+      }
     }
   }
 }
@@ -544,8 +550,10 @@ void interpolate(lite::Tensor* X,
   int out_w = Out->dims()[3];
   int spatial_in = in_h * in_w;
   int spatial_out = out_h * out_w;
-  for (int i = 0; i < count; ++i) {
-    if ("Bilinear" == interpolate_type) {
+
+  if ("Bilinear" == interpolate_type) {
+#pragma omp parallel for
+    for (int i = 0; i < count; ++i) {
       bilinear_interp(din + spatial_in * i,
                       in_w,
                       in_h,
@@ -555,7 +563,10 @@ void interpolate(lite::Tensor* X,
                       1.f / width_scale,
                       1.f / height_scale,
                       with_align);
-    } else if ("Nearest" == interpolate_type) {
+    }
+  } else if ("Nearest" == interpolate_type) {
+#pragma omp parallel for
+    for (int i = 0; i < count; ++i) {
       nearest_interp(din + spatial_in * i,
                      in_w,
                      in_h,

lite/kernels/arm/CMakeLists.txt

@@ -41,6 +41,7 @@ add_kernel(affine_channel_compute_arm ARM basic SRCS affine_channel_compute.cc D
 add_kernel(range_compute_arm ARM basic SRCS range_compute.cc DEPS ${lite_kernel_deps} math_arm)
 add_kernel(dropout_compute_arm ARM basic SRCS dropout_compute.cc DEPS ${lite_kernel_deps} math_arm)
 add_kernel(layout_compute_arm ARM basic SRCS layout_compute.cc DEPS ${lite_kernel_deps} math_arm)
+add_kernel(instance_norm_compute_arm ARM basic SRCS instance_norm_compute.cc DEPS ${lite_kernel_deps} math_arm)
 
 ## 2.other basic kernels: basic kernels that not used in basic models
 add_kernel(negative_compute_arm ARM extra SRCS negative_compute.cc DEPS ${lite_kernel_deps} math_arm)

lite/kernels/arm/instance_norm_compute.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 "lite/kernels/arm/instance_norm_compute.h"
+#include "lite/backends/arm/math/funcs.h"
+#include "lite/core/op_registry.h"
+#include "lite/core/type_system.h"
+
+namespace paddle {
+namespace lite {
+namespace kernels {
+namespace arm {
+
+void InstanceNormCompute::PrepareForRun() {}
+
+void InstanceNormCompute::Run() {
+  auto& param = this->Param<param_t>();
+  const float* in = param.x->data<float>();
+  const float* scale = param.scale->data<float>();
+  const float* bias = param.bias->data<float>();
+  float* out = param.out->mutable_data<float>();
+  float* saved_mean = param.saved_mean->mutable_data<float>();
+  float* saved_variance = param.saved_variance->mutable_data<float>();
+  float epsilon = param.epsilon;
+
+  int n = param.x->dims()[0];
+  int c = param.x->dims()[1];
+  int nc = n * c;
+  int height = param.x->dims()[2];
+  int width = param.x->dims()[3];
+  int spatial_size = height * width;
+// compute saved_mean and saved_variance
+#pragma omp parallel for
+  for (int i = 0; i < nc; ++i) {
+    const float* in_p = in + i * spatial_size;
+    float sum_spatial = 0.f;
+    float summ_spatial = 0.f;
+    for (int h = 0; h < height; ++h) {
+      int w = width;
+      float32x4_t sum0 = vdupq_n_f32(0.f);
+      float32x4_t sum1 = vdupq_n_f32(0.f);
+      float32x4_t sum2 = vdupq_n_f32(0.f);
+      float32x4_t sum3 = vdupq_n_f32(0.f);
+      float32x4_t summ0 = vdupq_n_f32(0.f);
+      float32x4_t summ1 = vdupq_n_f32(0.f);
+      float32x4_t summ2 = vdupq_n_f32(0.f);
+      float32x4_t summ3 = vdupq_n_f32(0.f);
+      float32x4_t in0, in1, in2, in3;
+      for (; w > 15; w -= 16) {
+        in0 = vld1q_f32(in_p);
+        in1 = vld1q_f32(in_p + 4);
+        in2 = vld1q_f32(in_p + 8);
+        in3 = vld1q_f32(in_p + 12);
+        sum0 = vaddq_f32(sum0, in0);
+        sum1 = vaddq_f32(sum1, in1);
+        summ0 = vmlaq_f32(summ0, in0, in0);
+        summ1 = vmlaq_f32(summ1, in1, in1);
+        sum2 = vaddq_f32(sum2, in2);
+        sum3 = vaddq_f32(sum3, in3);
+        summ2 = vmlaq_f32(summ2, in2, in2);
+        summ3 = vmlaq_f32(summ3, in3, in3);
+        in_p += 16;
+      }
+      for (; w > 7; w -= 8) {
+        in0 = vld1q_f32(in_p);
+        in1 = vld1q_f32(in_p + 4);
+        sum0 = vaddq_f32(sum0, in0);
+        sum1 = vaddq_f32(sum1, in1);
+        summ0 = vmlaq_f32(summ0, in0, in0);
+        summ1 = vmlaq_f32(summ1, in1, in1);
+        in_p += 8;
+      }
+      for (; w > 3; w -= 4) {
+        in0 = vld1q_f32(in_p);
+        sum0 = vaddq_f32(sum0, in0);
+        summ0 = vmlaq_f32(summ0, in0, in0);
+        in_p += 4;
+      }
+      float sum = 0.f;
+      float summ = 0.f;
+      for (; w > 0; w--) {
+        sum += *in_p;
+        summ += (*in_p) * (*in_p);
+        in_p++;
+      }
+      sum0 = vaddq_f32(sum0, sum1);
+      sum2 = vaddq_f32(sum2, sum3);
+      summ0 = vaddq_f32(summ0, summ1);
+      summ2 = vaddq_f32(summ2, summ3);
+      sum0 = vaddq_f32(sum0, sum2);
+      summ0 = vaddq_f32(summ0, summ2);
+      float32x2_t sum_low = vpadd_f32(vget_low_f32(sum0), vget_high_f32(sum0));
+      float32x2_t sum_high =
+          vpadd_f32(vget_low_f32(summ0), vget_high_f32(summ0));
+      float32x2_t sum_mix = vpadd_f32(sum_low, sum_high);
+      sum += vget_lane_f32(sum_mix, 0);
+      summ += vget_lane_f32(sum_mix, 1);
+      sum_spatial += sum;
+      summ_spatial += summ;
+    }
+    float mean = sum_spatial / spatial_size;
+    // float variance = summ / spatial_size - mean * mean;
+    // the flolowing code has higher precision than above comment code
+    float variance = (summ_spatial - mean * mean * spatial_size) / spatial_size;
+    float std = 1.f / sqrtf(variance + epsilon);
+
+    saved_mean[i] = mean;
+    saved_variance[i] = std;
+  }
+// compute instance_norm result: out = scale * (in - mean) / std + bias
+#pragma omp parallel for
+  for (int i = 0; i < nc; ++i) {
+    const float* in_p = in + i * spatial_size;
+    float* out_p = out + i * spatial_size;
+    int j = spatial_size;
+    const float sstd_val = scale[i % c] * saved_variance[i];
+    const float bias_val = bias[i % c];
+    const float mean_val = saved_mean[i];
+    const float32x4_t vsstd = vdupq_n_f32(sstd_val);
+    const float32x4_t vbias = vdupq_n_f32(bias_val);
+    const float32x4_t vmean = vdupq_n_f32(mean_val);
+    float32x4_t in0, in1, submean0, submean1, out0, out1;
+    for (; j > 7; j -= 8) {
+      in0 = vld1q_f32(in_p);
+      in1 = vld1q_f32(in_p + 4);
+      submean0 = vsubq_f32(in0, vmean);
+      submean1 = vsubq_f32(in1, vmean);
+      out0 = vmlaq_f32(vbias, submean0, vsstd);
+      out1 = vmlaq_f32(vbias, submean1, vsstd);
+      vst1q_f32(out_p, out0);
+      vst1q_f32(out_p + 4, out1);
+      in_p += 8;
+      out_p += 8;
+    }
+    for (; j > 3; j -= 4) {
+      in0 = vld1q_f32(in_p);
+      submean0 = vsubq_f32(in0, vmean);
+      out0 = vmlaq_f32(vbias, submean0, vsstd);
+      vst1q_f32(out_p, out0);
+      in_p += 4;
+      out_p += 4;
+    }
+    for (; j > 0; j--) {
+      *out_p = (*in_p - mean_val) * sstd_val + bias_val;
+      in_p++;
+      out_p++;
+    }
+  }
+}
+
+}  // namespace arm
+}  // namespace kernels
+}  // namespace lite
+}  // namespace paddle
+
+REGISTER_LITE_KERNEL(instance_norm,
+                     kARM,
+                     kFloat,
+                     kNCHW,
+                     paddle::lite::kernels::arm::InstanceNormCompute,
+                     def)
+    .BindInput("X", {LiteType::GetTensorTy(TARGET(kARM))})
+    .BindInput("Scale", {LiteType::GetTensorTy(TARGET(kARM))})
+    .BindInput("Bias", {LiteType::GetTensorTy(TARGET(kARM))})
+    .BindOutput("Y", {LiteType::GetTensorTy(TARGET(kARM))})
+    .BindOutput("SavedMean", {LiteType::GetTensorTy(TARGET(kARM))})
+    .BindOutput("SavedVariance", {LiteType::GetTensorTy(TARGET(kARM))})
+    .Finalize();

lite/kernels/arm/instance_norm_compute.h

+// 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.
+
+#pragma once
+#include "lite/core/kernel.h"
+#include "lite/core/op_registry.h"
+
+namespace paddle {
+namespace lite {
+namespace kernels {
+namespace arm {
+
+class InstanceNormCompute : public KernelLite<TARGET(kARM), PRECISION(kFloat)> {
+ public:
+  using param_t = operators::InstanceNormParam;
+
+  void PrepareForRun() override;
+
+  void Run() override;
+
+  virtual ~InstanceNormCompute() = default;
+
+ private:
+};
+
+}  // namespace arm
+}  // namespace kernels
+}  // namespace lite
+}  // namespace paddle

lite/operators/CMakeLists.txt

@@ -47,6 +47,7 @@ add_operator(fusion_elementwise_activation_ops basic SRCS fusion_elementwise_act
 add_operator(io_copy_once_op basic SRCS io_copy_once_op.cc DEPS io_copy_op ${op_DEPS})
 add_operator(dropout_op basic SRCS dropout_op.cc DEPS ${op_DEPS})
 add_operator(layout_op basic SRCS layout_op.cc DEPS ${op_DEPS})
+add_operator(instance_norm_op basic SRCS instance_norm_op.cc DEPS ${op_DEPS})
 add_operator(graph_op basic SRCS graph_op.cc DEPS ${op_DEPS})
 
 # 2.basic ops not used in basic models

lite/operators/instance_norm_op.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 "lite/operators/instance_norm_op.h"
+#include <string>
+#include <vector>
+#include "lite/core/op_lite.h"
+#include "lite/core/op_registry.h"
+#include "lite/core/tensor.h"
+
+namespace paddle {
+namespace lite {
+namespace operators {
+
+bool InstanceNormOp::CheckShape() const {
+  CHECK_OR_FALSE(param_.x);
+  CHECK_OR_FALSE(param_.scale);
+  CHECK_OR_FALSE(param_.bias);
+  CHECK_OR_FALSE(param_.out);
+  CHECK_OR_FALSE(param_.saved_mean);
+  CHECK_OR_FALSE(param_.saved_variance);
+  auto x_dims = param_.x->dims();
+  auto scale_dims = param_.scale->dims();
+  auto bias_dims = param_.bias->dims();
+  CHECK(x_dims.size() >= 2 && x_dims.size() <= 5)
+      << "Input X must have 2 to 5 dimensions.";
+  CHECK_EQ(scale_dims.size(), 1UL) << "Input Scale must have 1 dimensions.";
+  CHECK_EQ(bias_dims.size(), 1UL) << "Input Bias must have 1 dimensions.";
+  CHECK_GT(param_.epsilon, 0.f) << "epsilon should be greater than 0.f";
+  CHECK_LT(param_.epsilon, 0.01f) << "epsilon should be less than 0.01f";
+  return true;
+}
+
+bool InstanceNormOp::InferShape() const {
+  auto x_dims = param_.x->dims();
+  int64_t batch_size = x_dims[0];
+  int64_t channel_size = x_dims[1];
+  param_.saved_mean->Resize({batch_size * channel_size});
+  param_.saved_variance->Resize({batch_size * channel_size});
+  param_.out->Resize(x_dims);
+  return true;
+}
+
+bool InstanceNormOp::AttachImpl(const cpp::OpDesc& op_desc,
+                                lite::Scope* scope) {
+  param_.x = scope->FindVar(op_desc.Input("X").front())->GetMutable<Tensor>();
+  param_.scale =
+      scope->FindVar(op_desc.Input("Scale").front())->GetMutable<Tensor>();
+  param_.bias =
+      scope->FindVar(op_desc.Input("Bias").front())->GetMutable<Tensor>();
+  param_.saved_mean =
+      scope->FindVar(op_desc.Output("SavedMean").front())->GetMutable<Tensor>();
+  param_.saved_variance =
+      scope->FindVar(op_desc.Output("SavedVariance").front())
+          ->GetMutable<Tensor>();
+  param_.out =
+      scope->FindVar(op_desc.Output("Y").front())->GetMutable<Tensor>();
+  param_.epsilon = op_desc.GetAttr<float>("epsilon");
+  return true;
+}
+
+} /* namespace operators */
+} /* namespace lite */
+} /* namespace paddle */
+
+REGISTER_LITE_OP(instance_norm, paddle::lite::operators::InstanceNormOp);

lite/operators/instance_norm_op.h

+// 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.
+
+#pragma once
+#include <string>
+#include <vector>
+#include "lite/core/op_lite.h"
+#include "lite/core/scope.h"
+#include "lite/utils/all.h"
+
+namespace paddle {
+namespace lite {
+namespace operators {
+
+class InstanceNormOp : public OpLite {
+ public:
+  InstanceNormOp() {}
+
+  explicit InstanceNormOp(const std::string &op_type) : OpLite(op_type) {}
+
+  bool CheckShape() const override;
+
+  bool InferShape() 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 "instance_norm"; }
+
+ private:
+  mutable InstanceNormParam param_;
+};
+
+} /* namespace operators */
+} /* namespace lite */
+} /* namespace paddle */

lite/operators/op_params.h

@@ -1090,6 +1090,17 @@ struct CollectFpnProposalsParam {
   int post_nms_topN{};
 };
 
+/// --------------------- instance_norm operators --------------------
+struct InstanceNormParam {
+  lite::Tensor* x{};
+  lite::Tensor* out{};
+  lite::Tensor* bias{};
+  lite::Tensor* scale{};
+  lite::Tensor* saved_mean{};
+  lite::Tensor* saved_variance{};
+  float epsilon;
+};
+
 }  // namespace operators
 }  // namespace lite
 }  // namespace paddle

lite/tests/kernels/CMakeLists.txt

@@ -14,6 +14,7 @@ if((NOT LITE_WITH_OPENCL AND NOT LITE_WITH_FPGA AND NOT LITE_WITH_XPU) AND (LITE
     lite_cc_test(test_kernel_conv2d_transpose_compute SRCS conv2d_transpose_compute_test.cc DEPS arena_framework ${x86_kernels} ${arm_kernels} ${lite_ops} ${host_kernels})
     lite_cc_test(test_kernel_norm_compute SRCS norm_compute_test.cc DEPS arena_framework ${x86_kernels} ${arm_kernels} ${lite_ops} ${host_kernels})
     lite_cc_test(test_kernel_cast_compute SRCS cast_compute_test.cc DEPS arena_framework ${x86_kernels} ${arm_kernels} ${lite_ops} ${host_kernels})
+    lite_cc_test(test_kernel_instance_norm_compute SRCS instance_norm_compute_test.cc DEPS arena_framework ${x86_kernels} ${arm_kernels} ${lite_ops} ${host_kernels})
     #lite_cc_test(test_kernel_sequence_softmax_compute SRCS sequence_softmax_compute_test.cc DEPS arena_framework ${x86_kernels} ${arm_kernels} ${lite_ops} ${host_kernels})
     #lite_cc_test(test_kernel_im2sequence_compute SRCS im2sequence_compute_test.cc DEPS arena_framework ${x86_kernels} ${arm_kernels} ${lite_ops} ${host_kernels})
     #lite_cc_test(test_kernel_compare_compute SRCS compare_compute_test.cc DEPS arena_framework ${x86_kernels} ${arm_kernels} ${lite_ops} ${host_kernels})

lite/tests/kernels/instance_norm_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"
+#include "lite/tests/utils/fill_data.h"
+
+namespace paddle {
+namespace lite {
+
+class InstanceNormComputeTest : public arena::TestCase {
+ protected:
+  // common attributes for this op.
+  std::string input_ = "x";
+  std::string output_ = "y";
+  std::string saved_mean_ = "saved_mean";
+  std::string saved_variance_ = "saved_variance";
+  std::string scale_ = "scale";
+  std::string bias_ = "bias";
+
+  DDim dims_{{4, 5, 19, 19}};
+  float epsilon_ = 1e-5f;
+
+ public:
+  InstanceNormComputeTest(const Place& place,
+                          const std::string& alias,
+                          DDim dims,
+                          float epsilon)
+      : TestCase(place, alias), dims_(dims), epsilon_(epsilon) {}
+
+  void RunBaseline(Scope* scope) override {
+    auto x = scope->FindTensor(input_);
+    auto scale = scope->FindTensor(scale_);
+    auto bias = scope->FindTensor(bias_);
+    auto out = scope->NewTensor(output_);
+    auto saved_mean = scope->NewTensor(saved_mean_);
+    auto saved_variance = scope->NewTensor(saved_variance_);
+    CHECK(out);
+    CHECK(saved_mean);
+    CHECK(saved_variance);
+    DDim saved_dim({dims_[0] * dims_[1]});
+    out->Resize(dims_);
+    saved_mean->Resize(saved_dim);
+    saved_variance->Resize(saved_dim);
+
+    auto x_data = x->data<float>();
+    auto scale_data = scale->data<float>();
+    auto bias_data = bias->data<float>();
+    auto out_data = out->mutable_data<float>();
+    auto saved_mean_data = saved_mean->mutable_data<float>();
+    auto saved_variance_data = saved_variance->mutable_data<float>();
+
+    int n = x->dims()[0];
+    int c = x->dims()[1];
+    int spatial_size = x->dims()[2] * x->dims()[3];
+
+    // compute mean
+    for (int i = 0; i < n * c; ++i) {
+      const float* x_ptr = x_data + i * spatial_size;
+      float sum = 0.f;
+      for (int j = 0; j < spatial_size; ++j) {
+        sum += x_ptr[j];
+      }
+      saved_mean_data[i] = sum / spatial_size;
+    }
+    // compute variance
+    for (int i = 0; i < n * c; ++i) {
+      const float* x_ptr = x_data + i * spatial_size;
+      float sum = 0.f;
+      for (int j = 0; j < spatial_size; ++j) {
+        sum +=
+            (x_ptr[j] - saved_mean_data[i]) * (x_ptr[j] - saved_mean_data[i]);
+      }
+      saved_variance_data[i] = 1.f / sqrtf(sum / spatial_size + epsilon_);
+    }
+    // compute out
+    for (int i = 0; i < n * c; ++i) {
+      const float* x_ptr = x_data + i * spatial_size;
+      float* out_ptr = out_data + i * spatial_size;
+      float scale_val = scale_data[i % c];
+      float bias_val = bias_data[i % c];
+      for (int j = 0; j < spatial_size; ++j) {
+        out_ptr[j] = scale_val * (x_ptr[j] - saved_mean_data[i]) *
+                         saved_variance_data[i] +
+                     bias_val;
+      }
+    }
+  }
+
+  void PrepareOpDesc(cpp::OpDesc* op_desc) {
+    op_desc->SetType("instance_norm");
+    op_desc->SetInput("X", {input_});
+    op_desc->SetInput("Bias", {bias_});
+    op_desc->SetInput("Scale", {scale_});
+    op_desc->SetOutput("Y", {output_});
+    op_desc->SetOutput("SavedMean", {saved_mean_});
+    op_desc->SetOutput("SavedVariance", {saved_variance_});
+    op_desc->SetAttr("epsilon", epsilon_);
+  }
+
+  void PrepareData() override {
+    std::vector<float> din(dims_.production());
+    fill_data_rand(din.data(), -1.f, 1.f, dims_.production());
+
+    DDim scale_dim{{dims_[1]}};
+    std::vector<float> scale(scale_dim.production());
+    fill_data_rand(scale.data(), -1.f, 1.f, scale_dim.production());
+
+    std::vector<float> bias(scale_dim.production());
+    fill_data_rand(bias.data(), -1.f, 1.f, scale_dim.production());
+
+    SetCommonTensor(input_, dims_, din.data());
+    SetCommonTensor(scale_, scale_dim, scale.data());
+    SetCommonTensor(bias_, scale_dim, bias.data());
+  }
+};
+
+void test_instance_norm(Place place) {
+  for (auto& n : {1, 3, 16}) {
+    for (auto& c : {1, 4, 16}) {
+      for (auto& h : {1, 16, 33, 56}) {
+        for (auto& w : {1, 17, 34, 55}) {
+          DDim dim_in({n, c, h, w});
+          float epsilon = 1e-5f;
+          std::unique_ptr<arena::TestCase> tester(
+              new InstanceNormComputeTest(place, "def", dim_in, epsilon));
+#ifdef LITE_WITH_ARM
+          auto& ctx = tester->context()->As<ARMContext>();
+          ctx.SetRunMode(lite_api::LITE_POWER_HIGH, 4);
+#endif
+          arena::Arena arena(std::move(tester), place, 6e-5);
+          if (!arena.TestPrecision()) {
+            LOG(ERROR) << "run n: " << n << ", c: " << c << ", h: " << h
+                       << ", w: " << w;
+            return;
+          }
+        }
+      }
+    }
+  }
+}
+
+TEST(InstanceNorm, precision) {
+#ifdef LITE_WITH_ARM
+  Place place(TARGET(kARM));
+  test_instance_norm(place);
+#endif
+}
+
+}  // namespace lite
+}  // namespace paddle