[onert] Introduce DepthwiseConvOp in cpu kernel (#13574)

This commit introduces DepthwiseConvOp in cpu kernel.
This kernel uses depthwise_conv and bias_op eigen functions.

ONE-DCO-1.0-Signed-off-by: Jiyoung Yun <[email protected]>

compute/cker/include/cker/operation/DepthwiseConv.h

@@ -28,6 +28,8 @@
 #include "cker/operation/reference/integer_ops/DepthwiseConvUInt8.h"
 #include "cker/operation/reference/integer_ops/DepthwiseConvHybrid.h"
 #include "cker/CpuBackendThreadpool.h"
+#include "cker/eigen/depthwise_conv_op.h"
+#include "cker/eigen/bias_op.h"
 
 namespace nnfw
 {
@@ -180,6 +182,46 @@ inline void DepthwiseConv(const DepthwiseConvParams &params, const Shape &input_
   cpu_backend_threadpool::Execute(tasks.size(), tasks.data(), ruy_context);
 }
 
+void DepthwiseConvOp(const DepthwiseConvParams &params, const Shape &input_shape,
+                     const float *input_data, const Shape &filter_shape, const float *filter_data,
+                     const Shape &bias_shape, const float *bias_data, float *padded_filter_data,
+                     bool pad_filter, float *filter_buffers_data, const Shape &output_shape,
+                     float *output_data)
+{
+  if (params.stride_height != params.stride_width)
+    throw std::runtime_error("Not support different length strides");
+
+  if (params.dilation_height_factor != 1 || params.dilation_width_factor != 1)
+    throw std::runtime_error{"Not support dilation other than 1."};
+
+  const int batch = MatchingDim(input_shape, 0, output_shape, 0);
+  const int input_depth = input_shape.Dims(3);
+  const int output_depth = output_shape.Dims(3);
+  const int input_height = input_shape.Dims(1);
+  const int input_width = input_shape.Dims(2);
+  const int filter_height = filter_shape.Dims(1);
+  const int filter_width = filter_shape.Dims(2);
+  const int output_height = output_shape.Dims(1);
+  const int output_width = output_shape.Dims(2);
+  const int stride = params.stride_height;
+  const int depth_multiplier = params.depth_multiplier;
+  const int pad_height = params.padding_values.height;
+  const int pad_width = params.padding_values.width;
+  const float activation_min = params.float_activation_min;
+  const float activation_max = params.float_activation_max;
+
+  depthwise_conv_op::LaunchDepthwiseConvOp<Eigen::ThreadPoolDevice, float>()(
+    batch, input_height, input_width, input_depth, filter_height, filter_width, depth_multiplier,
+    stride, pad_height, pad_width, output_height, output_width, output_depth, input_data,
+    filter_data, padded_filter_data, pad_filter, filter_buffers_data, output_data);
+
+  if (bias_data != nullptr)
+  {
+    bias_op::biasHelper<float>(bias_shape, bias_data, output_shape, output_data, activation_min,
+                               activation_max);
+  }
+}
+
 } // namespace cker
 } // namespace nnfw
 

compute/cker/src/DepthwiseConv.test.cc

@@ -0,0 +1,303 @@
+/*
+ * Copyright (c) 2024 Samsung Electronics Co., Ltd. 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 <cker/eigen/EigenSupport.h>
+#include <cker/operation/DepthwiseConv.h>
+
+#include <gtest/gtest.h>
+#include <vector>
+
+namespace
+{
+
+template <typename T> class DepthwiseConvVerifier
+{
+public:
+  DepthwiseConvVerifier() = default;
+
+  void prepare(const nnfw::cker::Shape &output_shape, const nnfw::cker::Shape &filter_shape)
+  {
+    const int k_packet_size = nnfw::cker::eigen_support::kPacketSize<T>();
+    const int batch = output_shape.Dims(0);
+    const int out_depth = output_shape.Dims(3);
+    const int filter_rows = filter_shape.Dims(1);
+    const int filter_cols = filter_shape.Dims(2);
+    const int filter_spatial_size = filter_rows * filter_cols;
+    const int padded_filter_inner_dim_size =
+      ((out_depth + k_packet_size - 1) / k_packet_size) * k_packet_size;
+
+    _use_padded_filter = (out_depth % k_packet_size) == 0 ? false : true;
+    {
+      nnfw::cker::Shape padded_filter_shape(
+        {batch, filter_spatial_size, padded_filter_inner_dim_size});
+      _padded_filter.resize(padded_filter_shape.FlatSize());
+    }
+
+    {
+      // NOTE The Eigen library uses both main thread as well as a thread pool.
+      // Therefore, it needs to add an additional memory buffer for main thread.
+      const int thread_count = nnfw::cker::eigen_support::getThreadCount() + 1;
+
+      nnfw::cker::Shape filter_buffer_shape(
+        {thread_count, filter_spatial_size, padded_filter_inner_dim_size});
+      _filter_buffers.resize(filter_buffer_shape.FlatSize());
+    }
+  }
+
+  void run(const nnfw::cker::DepthwiseConvParams &params, const nnfw::cker::Shape &input_shape,
+           const T *input_data, const nnfw::cker::Shape &filter_shape, const T *filter_data,
+           const nnfw::cker::Shape &bias_shape, const T *bias_data,
+           const nnfw::cker::Shape &output_shape, const T *expected)
+  {
+    std::vector<T> output(output_shape.FlatSize());
+    nnfw::cker::DepthwiseConvOp(params, input_shape, input_data, filter_shape, filter_data,
+                                bias_shape, bias_data, _padded_filter.data(), _use_padded_filter,
+                                _filter_buffers.data(), output_shape, output.data());
+
+    for (size_t i = 0; i < output.size(); ++i)
+      EXPECT_NEAR(output[i], expected[i], 1e-3f);
+  }
+
+  void checkException(const nnfw::cker::DepthwiseConvParams &params,
+                      const nnfw::cker::Shape &input_shape, const T *input_data,
+                      const nnfw::cker::Shape &filter_shape, const T *filter_data,
+                      const nnfw::cker::Shape &bias_shape, const T *bias_data,
+                      const nnfw::cker::Shape &output_shape, const T *expected)
+  {
+    std::vector<T> output(output_shape.FlatSize());
+    EXPECT_ANY_THROW(
+      nnfw::cker::DepthwiseConvOp(params, input_shape, input_data, filter_shape, filter_data,
+                                  bias_shape, bias_data, _padded_filter.data(), _use_padded_filter,
+                                  _filter_buffers.data(), output_shape, output.data()));
+  }
+
+private:
+  bool _use_padded_filter;
+  std::vector<T> _padded_filter;
+  std::vector<T> _filter_buffers;
+};
+
+} // namespace
+
+TEST(CKer_Operation, DepthwiseConv)
+{
+  {
+    nnfw::cker::DepthwiseConvParams params{};
+    params.padding_type = nnfw::cker::PaddingType::kValid;
+    params.padding_values.width = 0;
+    params.padding_values.height = 0;
+    params.stride_width = 1;
+    params.stride_height = 1;
+    params.dilation_width_factor = 1;
+    params.dilation_height_factor = 1;
+    params.depth_multiplier = 1;
+    params.float_activation_min = std::numeric_limits<float>::lowest();
+    params.float_activation_max = std::numeric_limits<float>::max();
+
+    nnfw::cker::Shape input_shape{1, 3, 2, 2}; // n, h, w, c
+    std::vector<float> input = {1.0, 2.0, 7.0, 8.0, 3.0, 4.0, 9.0, 10.0, 5.0, 6.0, 11.0, 12.0};
+    nnfw::cker::Shape filter_shape{1, 2, 2, 2}; // 1, h, w, c
+    std::vector<float> filter = {1.0, 2.0, 3.0, 4.0, -9.0, 10.0, -11.0, 12.0};
+    nnfw::cker::Shape bias_shape{2};
+    std::vector<float> bias = {0.0, 0.0};
+    nnfw::cker::Shape output_shape{1, 2, 1, 2}; // n, h, w, c
+    std::vector<float> expected = {-104., 196.0, -136.0, 252.0};
+
+    DepthwiseConvVerifier<float> verifier;
+    verifier.prepare(output_shape, filter_shape);
+    verifier.run(params, input_shape, input.data(), filter_shape, filter.data(), bias_shape,
+                 bias.data(), output_shape, expected.data());
+  }
+
+  // Pad
+  {
+    nnfw::cker::DepthwiseConvParams params{};
+    params.padding_type = nnfw::cker::PaddingType::kSame;
+    params.padding_values.width = 0;
+    params.padding_values.height = 1;
+    params.stride_width = 1;
+    params.stride_height = 1;
+    params.dilation_width_factor = 1;
+    params.dilation_height_factor = 1;
+    params.depth_multiplier = 1;
+    params.float_activation_min = std::numeric_limits<float>::lowest();
+    params.float_activation_max = std::numeric_limits<float>::max();
+
+    nnfw::cker::Shape input_shape{1, 2, 2, 2}; // n, h, w, c
+    std::vector<float> input = {0.0, 1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0};
+    nnfw::cker::Shape filter_shape{1, 3, 1, 2}; // 1, h, w, c
+    std::vector<float> filter = {0.0, 1.0, 2.0, 3.0, 4.0, 5.0};
+    nnfw::cker::Shape bias_shape{2};
+    std::vector<float> bias = {0.0, 0.0};
+    nnfw::cker::Shape output_shape{1, 2, 2, 2}; // n, h, w, c
+    std::vector<float> expected = {16.0, 28.0, 28.0, 44.0, 8.0, 16.0, 12.0, 24.0};
+
+    DepthwiseConvVerifier<float> verifier;
+    verifier.prepare(output_shape, filter_shape);
+    verifier.run(params, input_shape, input.data(), filter_shape, filter.data(), bias_shape,
+                 bias.data(), output_shape, expected.data());
+  }
+
+  // Bias
+  {
+    nnfw::cker::DepthwiseConvParams params{};
+    params.padding_type = nnfw::cker::PaddingType::kSame;
+    params.padding_values.width = 0;
+    params.padding_values.height = 1;
+    params.stride_width = 1;
+    params.stride_height = 1;
+    params.dilation_width_factor = 1;
+    params.dilation_height_factor = 1;
+    params.depth_multiplier = 1;
+    params.float_activation_min = std::numeric_limits<float>::lowest();
+    params.float_activation_max = std::numeric_limits<float>::max();
+
+    nnfw::cker::Shape input_shape{1, 2, 2, 2}; // n, h, w, c
+    std::vector<float> input = {0.0, 1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0};
+    nnfw::cker::Shape filter_shape{1, 3, 1, 2}; // 1, h, w, c
+    std::vector<float> filter = {0.0, 1.0, 2.0, 3.0, 4.0, 5.0};
+    nnfw::cker::Shape bias_shape{2};
+    std::vector<float> bias = {0.5, -0.5};
+    nnfw::cker::Shape output_shape{1, 2, 2, 2}; // n, h, w, c
+    std::vector<float> expected = {16.5, 27.5, 28.5, 43.5, 8.5, 15.5, 12.5, 23.5};
+
+    DepthwiseConvVerifier<float> verifier;
+    verifier.prepare(output_shape, filter_shape);
+    verifier.run(params, input_shape, input.data(), filter_shape, filter.data(), bias_shape,
+                 bias.data(), output_shape, expected.data());
+  }
+
+  // Depth Multiplier
+  {
+    nnfw::cker::DepthwiseConvParams params{};
+    params.padding_type = nnfw::cker::PaddingType::kSame;
+    params.padding_values.width = 0;
+    params.padding_values.height = 1;
+    params.stride_width = 1;
+    params.stride_height = 1;
+    params.dilation_width_factor = 1;
+    params.dilation_height_factor = 1;
+    params.depth_multiplier = 2;
+    params.float_activation_min = std::numeric_limits<float>::lowest();
+    params.float_activation_max = std::numeric_limits<float>::max();
+
+    nnfw::cker::Shape input_shape{1, 2, 2, 2}; // n, h, w, c
+    std::vector<float> input = {0.0, 1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0};
+    nnfw::cker::Shape filter_shape{1, 3, 1, 4}; // 1, h, w, c
+    std::vector<float> filter = {0.0, 1.0, 2.0, 3.0, 4.0, 5.0, -5.0, -4.0, -3.0, -2.0, -1.0, 0.0};
+    nnfw::cker::Shape bias_shape{4};
+    std::vector<float> bias = {0.5, -0.5, 0.3, -0.3};
+    nnfw::cker::Shape output_shape{1, 2, 2, 4}; // n, h, w, c
+    std::vector<float> expected = {-11.5, -8.5, -9.7,  -4.3,  -9.5, -2.5, -21.7, -12.3,
+                                   16.5,  19.5, -22.7, -17.3, 24.5, 31.5, -28.7, -19.3};
+
+    DepthwiseConvVerifier<float> verifier;
+    verifier.prepare(output_shape, filter_shape);
+    verifier.run(params, input_shape, input.data(), filter_shape, filter.data(), bias_shape,
+                 bias.data(), output_shape, expected.data());
+  }
+
+  // ReLU6
+  {
+    nnfw::cker::DepthwiseConvParams params{};
+    params.padding_type = nnfw::cker::PaddingType::kSame;
+    params.padding_values.width = 0;
+    params.padding_values.height = 1;
+    params.stride_width = 1;
+    params.stride_height = 1;
+    params.dilation_width_factor = 1;
+    params.dilation_height_factor = 1;
+    params.depth_multiplier = 1;
+    params.float_activation_min = 0.0;
+    params.float_activation_max = 6.0;
+
+    nnfw::cker::Shape input_shape{1, 2, 2, 2}; // n, h, w, c
+    std::vector<float> input = {0.0, 1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0};
+    nnfw::cker::Shape filter_shape{1, 3, 1, 2}; // 1, h, w, c
+    std::vector<float> filter = {0.0, 1.0, 2.0, 3.0, 4.0, 5.0};
+    nnfw::cker::Shape bias_shape{2};
+    std::vector<float> bias = {0.5, -0.5};
+    nnfw::cker::Shape output_shape{1, 2, 2, 2}; // n, h, w, c
+    std::vector<float> expected = {6.0, 6.0, 6.0, 6.0, 6.0, 6.0, 6.0, 6.0};
+
+    DepthwiseConvVerifier<float> verifier;
+    verifier.prepare(output_shape, filter_shape);
+    verifier.run(params, input_shape, input.data(), filter_shape, filter.data(), bias_shape,
+                 bias.data(), output_shape, expected.data());
+  }
+
+  // No bias
+  {
+    nnfw::cker::DepthwiseConvParams params{};
+    params.padding_type = nnfw::cker::PaddingType::kSame;
+    params.padding_values.width = 0;
+    params.padding_values.height = 1;
+    params.stride_width = 1;
+    params.stride_height = 1;
+    params.dilation_width_factor = 1;
+    params.dilation_height_factor = 1;
+    params.depth_multiplier = 1;
+    params.float_activation_min = std::numeric_limits<float>::lowest();
+    params.float_activation_max = std::numeric_limits<float>::max();
+
+    nnfw::cker::Shape input_shape{1, 2, 2, 2}; // n, h, w, c
+    std::vector<float> input = {0.0, 1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0};
+    nnfw::cker::Shape filter_shape{1, 3, 1, 2}; // 1, h, w, c
+    std::vector<float> filter = {0.0, 1.0, 2.0, 3.0, 4.0, 5.0};
+    nnfw::cker::Shape bias_shape{2};
+    nnfw::cker::Shape output_shape{1, 2, 2, 2}; // n, h, w, c
+    std::vector<float> expected = {16.0, 28.0, 28.0, 44.0, 8.0, 16.0, 12.0, 24.0};
+
+    DepthwiseConvVerifier<float> verifier;
+    verifier.prepare(output_shape, filter_shape);
+    verifier.run(params, input_shape, input.data(), filter_shape, filter.data(), bias_shape,
+                 nullptr, output_shape, expected.data());
+  }
+}
+
+TEST(CKer_Operation, neg_DepthwiseConv)
+{
+  // Not supported Dilation
+  {
+    nnfw::cker::DepthwiseConvParams params{};
+    params.padding_type = nnfw::cker::PaddingType::kSame;
+    params.padding_values.width = 0;
+    params.padding_values.height = 1;
+    params.stride_width = 1;
+    params.stride_height = 1;
+    params.dilation_width_factor = 2;
+    params.dilation_height_factor = 2;
+    params.depth_multiplier = 1;
+    params.float_activation_min = std::numeric_limits<float>::lowest();
+    params.float_activation_max = std::numeric_limits<float>::max();
+
+    nnfw::cker::Shape input_shape{1, 6, 6, 1}; // n, h, w, c
+    std::vector<float> input = {0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
+                                0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0,
+                                0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0};
+    nnfw::cker::Shape filter_shape{1, 2, 2, 1}; // 1, h, w, c
+    std::vector<float> filter = {1.0, 2.0, 3.0, 4.0};
+    nnfw::cker::Shape bias_shape{1};
+    std::vector<float> bias = {0.0};
+    nnfw::cker::Shape output_shape{1, 3, 3, 1}; // n, h, w, c
+    std::vector<float> expected = {4.0, 0.0, 3.0, 0.0, 0.0, 0.0, 2.0, 0.0, 1.0};
+
+    DepthwiseConvVerifier<float> verifier;
+    verifier.prepare(output_shape, filter_shape);
+    verifier.checkException(params, input_shape, input.data(), filter_shape, filter.data(),
+                            bias_shape, bias.data(), output_shape, expected.data());
+  }
+}