commit 44b4e974590f1a6a07b235f203006cc9010b37e8
author George Wort <george.wort@arm.com> Tue Jan 08 11:41:54 2019 +0000
committer George Wort <george.wort@arm.com> Wed Jan 16 15:55:43 2019 +0000
tree 1f7f76712847e9b7269bc56f972006dd9902ea3c
parent f63885bc445af1329e6a5c44d94b5c5d78146b2c

COMPMID-1794: Add support for NHWC in CLROIAlignLayer

Change-Id: If1df8f6c0549c986e607cbceb0977c80b2891b75
Reviewed-on: https://review.mlplatform.org/493
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Isabella Gottardi <isabella.gottardi@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>

src/core/CL/kernels/CLROIAlignLayerKernel.cpp

diff --git a/src/core/CL/kernels/CLROIAlignLayerKernel.cpp b/src/core/CL/kernels/CLROIAlignLayerKernel.cpp
index 325eeb2..66d2623 100644
--- a/src/core/CL/kernels/CLROIAlignLayerKernel.cpp
+++ b/src/core/CL/kernels/CLROIAlignLayerKernel.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018 ARM Limited.
+ * Copyright (c) 2018-2019 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -34,6 +34,9 @@
 #include "arm_compute/core/TensorInfo.h"
 #include "arm_compute/core/Utils.h"
 #include "arm_compute/core/Window.h"
+#include "arm_compute/core/utils/misc/ShapeCalculator.h"
+
+using namespace arm_compute::misc::shape_calculator;
 
 namespace arm_compute
 {
@@ -47,18 +50,15 @@
     ARM_COMPUTE_RETURN_ERROR_ON(rois->num_dimensions() > 2);
     ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(input);
     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::F32, DataType::F16);
-    ARM_COMPUTE_RETURN_ERROR_ON_DATA_LAYOUT_NOT_IN(input, DataLayout::NCHW);
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_LAYOUT_NOT_IN(input, DataLayout::NHWC, DataLayout::NCHW);
     ARM_COMPUTE_RETURN_ERROR_ON((pool_info.pooled_width() == 0) || (pool_info.pooled_height() == 0));
 
     if(output->total_size() != 0)
     {
         ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
         ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_LAYOUT(input, output);
-        ARM_COMPUTE_RETURN_ERROR_ON((output->dimension(0) != pool_info.pooled_width()) || (output->dimension(1) != pool_info.pooled_height()));
-        ARM_COMPUTE_RETURN_ERROR_ON(input->dimension(2) != output->dimension(2));
-        ARM_COMPUTE_RETURN_ERROR_ON(rois->dimension(1) != output->dimension(3));
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(compute_roi_align_shape(*input, *rois, pool_info), output->tensor_shape());
     }
-
     return Status{};
 }
 
@@ -67,8 +67,9 @@
     ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
 
     // Output auto inizialitation if not yet initialized
-    TensorShape output_shape(pool_info.pooled_width(), pool_info.pooled_height(), input->dimension(2), rois->dimension(1));
+    const TensorShape output_shape = compute_roi_align_shape(*input, *rois, pool_info);
     auto_init_if_empty((*output), output_shape, 1, input->data_type());
+    output->set_data_layout(input->data_layout());
 
     // Configure kernel window
     const unsigned int num_elems_processed_per_iteration = 1;
@@ -107,12 +108,13 @@
     CLBuildOptions build_opts;
     build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(input->info()->data_type()));
     build_opts.add_option("-DDATA_SIZE=" + get_data_size_from_data_type(input->info()->data_type()));
-    build_opts.add_option("-DMAX_DIM_X=" + support::cpp11::to_string(_input->info()->dimension(Window::DimX)));
-    build_opts.add_option("-DMAX_DIM_Y=" + support::cpp11::to_string(_input->info()->dimension(Window::DimY)));
-    build_opts.add_option("-DMAX_DIM_Z=" + support::cpp11::to_string(_input->info()->dimension(Window::DimZ)));
+    build_opts.add_option("-DMAX_DIM_X=" + support::cpp11::to_string(_input->info()->dimension(get_data_layout_dimension_index(input->info()->data_layout(), DataLayoutDimension::WIDTH))));
+    build_opts.add_option("-DMAX_DIM_Y=" + support::cpp11::to_string(_input->info()->dimension(get_data_layout_dimension_index(input->info()->data_layout(), DataLayoutDimension::HEIGHT))));
+    build_opts.add_option("-DMAX_DIM_Z=" + support::cpp11::to_string(_input->info()->dimension(get_data_layout_dimension_index(input->info()->data_layout(), DataLayoutDimension::CHANNEL))));
     build_opts.add_option("-DPOOLED_DIM_X=" + support::cpp11::to_string(pool_info.pooled_width()));
     build_opts.add_option("-DPOOLED_DIM_Y=" + support::cpp11::to_string(pool_info.pooled_height()));
     build_opts.add_option("-DSPATIAL_SCALE=" + float_to_string_with_full_precision(pool_info.spatial_scale()));
+    build_opts.add_option_if(input->info()->data_layout() == DataLayout::NHWC, "-DNHWC");
     build_opts.add_option_if(pool_info.sampling_ratio() > 0, "-DSAMPLING_RATIO=" + support::cpp11::to_string(pool_info.sampling_ratio()));
 
     // Create kernel
@@ -137,7 +139,7 @@
     Window slice_rois = slice;
     // Parallelize spatially and across the fourth dimension of the output tensor (also across ROITensor)
     slice_rois.set_dimension_step(Window::DimX, _rois->info()->dimension(0));
-    slice.set(Window::DimZ, window[3]);
+    slice.set(get_data_layout_dimension_index(_input->info()->data_layout(), DataLayoutDimension::CHANNEL), window[3]);
 
     // Set arguments
     unsigned int idx = 0;