1 files changed, 117 insertions, 0 deletions
diff --git a/libs/ARMComputeEx/src/core/CL/kernels/CLBatchToSpaceNDKernel.cpp b/libs/ARMComputeEx/src/core/CL/kernels/CLBatchToSpaceNDKernel.cpp
new file mode 100644
index 000000000..b0016d23c
--- /dev/null
+++ b/libs/ARMComputeEx/src/core/CL/kernels/CLBatchToSpaceNDKernel.cpp
@@ -0,0 +1,117 @@
+/*
+ * Copyright (c) 2018 Samsung Electronics Co., Ltd. All Rights Reserved
+ * Copyright (c) 2016-2018 ARM Limited.
+ *
+ * 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 "arm_compute/core/CL/kernels/CLBatchToSpaceNDKernel.h"
+
+#include "arm_compute/core/CL/CLHelpers.h"
+#include "arm_compute/core/CL/CLKernelLibraryEx.h"
+#include "arm_compute/core/CL/ICLTensor.h"
+
+using namespace arm_compute;
+
+namespace
+{
+Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output,
+                          const int32_t *block_size)
+{
+  ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::U8, DataType::QASYMM8,
+                                                DataType::S16, DataType::S32, DataType::F16,
+                                                DataType::F32);
+  ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::U8, DataType::QASYMM8,
+                                                DataType::S16, DataType::S32, DataType::F16,
+                                                DataType::F32);
+  ARM_COMPUTE_RETURN_ERROR_ON_MSG(block_size[0] >= 1 && block_size[1] >= 1,
+                                  "Block size should be greater than or equal to 1.");
+
+  ARM_COMPUTE_RETURN_ERROR_ON_MSG(input->dimension(2) == output->dimension(2),
+                                  "Input Depth should be equal to Output Depth");
+
+  ARM_COMPUTE_RETURN_ERROR_ON_MSG(
+      output->dimension(3) * block_size[0] * block_size[1] == input->dimension(3),
+      "Input batch should be equal to (output batch * block size[0] *block size[1])");
+
+  ARM_COMPUTE_RETURN_ERROR_ON_MSG(!(output->dimension(0) % block_size[1]) &&
+                                      !(output->dimension(1) % block_size[0]),
+                                  "Output height and width should be divisible by block size[0] "
+                                  "and block_size[1] respectively");
+
+  ARM_COMPUTE_RETURN_ERROR_ON_MSG((output->dimension(0) == input->dimension(0) * block_size[1]) &&
+                                      (output->dimension(1) == input->dimension(1) * block_size[0]),
+                                  "Output height and width should be equal to "
+                                  "input_height*blocksize[0] and input_width*blocksize[1] "
+                                  "respectively");
+
+  return Status{};
+}
+
+} // namespace
+
+CLBatchToSpaceNDKernel::CLBatchToSpaceNDKernel() : _input(nullptr), _output(nullptr) {}
+
+void CLBatchToSpaceNDKernel::configure(const ICLTensor *input, ICLTensor *output,
+                                       const int32_t *block_size)
+{
+
+  ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
+  ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), block_size));
+
+  _input = input;
+  _output = output;
+
+  // Set kernel build options
+  std::set<std::string> build_opts;
+  build_opts.emplace("-DDATA_TYPE=" + get_cl_type_from_data_type(input->info()->data_type()));
+  build_opts.emplace("-DBLOCK_SIZE0=" + support::cpp11::to_string(block_size[0]));
+  build_opts.emplace("-DBLOCK_SIZE1=" + support::cpp11::to_string(block_size[1]));
+  build_opts.emplace("-DBATCH_OUT=" + support::cpp11::to_string(output->info()->dimension(3)));
+  build_opts.emplace("-DDEPTH_OUT=" + support::cpp11::to_string(output->info()->dimension(2)));
+
+  // Create kernel
+  _kernel = static_cast<cl::Kernel>(
+      CLKernelLibraryEx::get().create_kernel("batch_to_space_nd", build_opts));
+
+  // Configure  kernel window
+  Window win = calculate_max_window(*output->info(), Steps());
+
+  Coordinates coord;
+  coord.set_num_dimensions(output->info()->num_dimensions());
+  output->info()->set_valid_region(ValidRegion(coord, output->info()->tensor_shape()));
+
+  ICLKernel::configure_internal(win);
+}
+
+void CLBatchToSpaceNDKernel::run(const Window &window, cl::CommandQueue &queue)
+{
+  ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
+  ARM_COMPUTE_ERROR_ON_MISMATCHING_WINDOWS(ICLKernel::window(), window);
+
+  Window slice_in = window.first_slice_window_4D().collapse(ICLKernel::window(), 2, 4);
+
+  // Setup output slice
+  Window slice_out(slice_in);
+  slice_out.set(Window::DimX, Window::Dimension(0, 0, 0));
+  slice_out.set(Window::DimY, Window::Dimension(0, 0, 0));
+  slice_out.set(Window::DimZ, Window::Dimension(0, 0, 0));
+  slice_out.set(3, Window::Dimension(0, 0, 0));
+
+  do
+  {
+    unsigned int idx = 0;
+    add_4D_tensor_argument(idx, _input, slice_out);
+    add_4D_tensor_argument(idx, _output, slice_in);
+    enqueue(queue, *this, slice_in);
+  } while (window.slide_window_slice_4D(slice_out) && window.slide_window_slice_4D(slice_in));
+}