COMPMID-791: Generic Depthwise Convolution Layer NEON QASYMM8

Change-Id: I33cf54e68f6c097ac58b6f16c3f9a720978f09cd
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/117289
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Michalis Spyrou <michalis.spyrou@arm.com>
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>

3 files changed, 28 insertions, 11 deletions

diff --git a/tests/datasets/DepthwiseConvolutionLayerDataset.h b/tests/datasets/DepthwiseConvolutionLayerDataset.h
index c5a9f96a2e..d3eb2c5d9e 100644
--- a/tests/datasets/DepthwiseConvolutionLayerDataset.h
+++ b/tests/datasets/DepthwiseConvolutionLayerDataset.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -118,7 +118,7 @@ class SmallDepthwiseConvolutionLayerDataset final : public DepthwiseConvolutionL
 public:
     SmallDepthwiseConvolutionLayerDataset()
     {
-        add_config(TensorShape(7U, 7U, 3U), TensorShape(3U, 3U, 3U), TensorShape(5U, 5U, 3U), PadStrideInfo(1, 1, 0, 0));
+        add_config(TensorShape(7U, 7U, 1U), TensorShape(3U, 3U, 1U), TensorShape(5U, 5U, 1U), PadStrideInfo(1, 1, 0, 0));
         add_config(TensorShape(23U, 27U, 5U), TensorShape(3U, 5U, 5U), TensorShape(11U, 23U, 5U), PadStrideInfo(2, 1, 0, 0));
         add_config(TensorShape(33U, 27U, 7U), TensorShape(7U, 3U, 7U), TensorShape(10U, 13U, 7U), PadStrideInfo(3, 2, 1, 0));
         add_config(TensorShape(33U, 27U, 11U), TensorShape(3U, 3U, 11U), TensorShape(31U, 14U, 11U), PadStrideInfo(1, 2, 0, 1));
diff --git a/tests/validation/NEON/DepthwiseConvolutionLayer.cpp b/tests/validation/NEON/DepthwiseConvolutionLayer.cpp
index e8c771595e..f8c04dab3e 100644
--- a/tests/validation/NEON/DepthwiseConvolutionLayer.cpp
+++ b/tests/validation/NEON/DepthwiseConvolutionLayer.cpp
@@ -128,9 +128,19 @@ TEST_SUITE_END()
 
 template <typename T>
 using NEDepthwiseConvolutionLayerQuantizedFixture3x3 = DepthwiseConvolutionLayerValidationQuantizedFixture<Tensor, Accessor, NEDepthwiseConvolutionLayer3x3, T>;
+template <typename T>
+using NEDepthwiseConvolutionLayerQuantizedFixture = DepthwiseConvolutionLayerValidationQuantizedFixture<Tensor, Accessor, NEDepthwiseConvolutionLayer, T>;
 
 TEST_SUITE(Quantized)
 TEST_SUITE(QASYMM8)
+TEST_SUITE(Generic)
+FIXTURE_DATA_TEST_CASE(RunSmall, NEDepthwiseConvolutionLayerQuantizedFixture<uint8_t>, framework::DatasetMode::PRECOMMIT, combine(combine(datasets::SmallDepthwiseConvolutionLayerDataset(),
+                       framework::dataset::make("DataType", DataType::QASYMM8)),
+                       framework::dataset::make("QuantizationInfo", { QuantizationInfo(0.5f, 10) })))
+{
+    validate(Accessor(_target), _reference, tolerance_qasymm8);
+}
+TEST_SUITE_END()
 TEST_SUITE(W3x3)
 FIXTURE_DATA_TEST_CASE(RunSmall, NEDepthwiseConvolutionLayerQuantizedFixture3x3<uint8_t>, framework::DatasetMode::PRECOMMIT, combine(combine(datasets::SmallDepthwiseConvolutionLayerDataset3x3(),
                        framework::dataset::make("DataType", DataType::QASYMM8)),
diff --git a/tests/validation/reference/DepthwiseConvolutionLayer.cpp b/tests/validation/reference/DepthwiseConvolutionLayer.cpp
index 6ca347f1d4..66e3a4b783 100644
--- a/tests/validation/reference/DepthwiseConvolutionLayer.cpp
+++ b/tests/validation/reference/DepthwiseConvolutionLayer.cpp
@@ -140,11 +140,18 @@ SimpleTensor<uint8_t> depthwise_convolution(const SimpleTensor<uint8_t> &src, co
     const int input_depth   = src.shape().z();
     const int num_batches   = src.shape().total_size() / (input_width * input_height * input_depth);
 
-    const int filter_half_size = filter_width / 2;
-    const int pad_x            = std::min(filter_half_size, static_cast<int>(conv_info.pad().first));
-    const int pad_y            = std::min(filter_half_size, static_cast<int>(conv_info.pad().second));
-    const int minimum_x        = -pad_x + filter_half_size;
-    const int minimum_y        = -pad_y + filter_half_size;
+    const int filter_half_width  = filter_width / 2;
+    const int filter_half_height = filter_height / 2;
+
+    const int pad_left   = std::min(static_cast<int>(conv_info.pad_left()), filter_half_width);
+    const int pad_top    = std::min(static_cast<int>(conv_info.pad_top()), filter_half_height);
+    const int pad_right  = std::min(static_cast<int>(conv_info.pad_right()), filter_half_width);
+    const int pad_bottom = std::min(static_cast<int>(conv_info.pad_bottom()), filter_half_height);
+
+    const int minimum_x = -pad_left + filter_half_width;
+    const int minimum_y = -pad_top + filter_half_height;
+    const int maximum_x = input_width + pad_left - filter_half_width + pad_right - filter_half_width;
+    const int maximum_y = input_height + pad_top - filter_half_height + pad_bottom - filter_half_height;
 
     int out_pos = 0;
     for(int r = 0; r < num_batches; ++r)
@@ -152,17 +159,17 @@ SimpleTensor<uint8_t> depthwise_convolution(const SimpleTensor<uint8_t> &src, co
         for(int z = 0; z < input_depth; ++z)
         {
             int32_t bias_val = *static_cast<const int32_t *>(biases(Coordinates(z)));
-            for(int y = minimum_y; y < input_height + pad_y - filter_half_size; y += conv_info.stride().second)
+            for(int y = minimum_y; y < minimum_y + maximum_y; y += conv_info.stride().second)
             {
-                for(int x = minimum_x; x < input_width + pad_x - filter_half_size; x += conv_info.stride().first)
+                for(int x = minimum_x; x < minimum_x + maximum_x; x += conv_info.stride().first)
                 {
                     Coordinates coords(x, y, z, r);
                     int         filter_offset = filter_plane * z;
 
                     int32_t val = 0;
-                    for(int j = y - filter_half_size; j <= (y + filter_half_size); ++j)
+                    for(int j = y - filter_half_height; j <= (y + filter_half_height); ++j)
                     {
-                        for(int i = x - filter_half_size; i <= (x + filter_half_size); ++i)
+                        for(int i = x - filter_half_width; i <= (x + filter_half_width); ++i)
                         {
                             coords.set(0, i);
                             coords.set(1, j);