COMPMID-1566: Add broadcast to CLArithmeticSubtraction

Change-Id: I05d21f9a92013ecfd1128d12cf1561cfd6e5c5e9
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/147983
Tested-by: bsgcomp <bsgcomp@arm.com>
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>

6 files changed, 44 insertions, 140 deletions

diff --git a/tests/validation/reference/ArithmeticAddition.cpp b/tests/validation/reference/ArithmeticOperations.cpp
index c68c6d4ed6..062be93b95 100644
--- a/tests/validation/reference/ArithmeticAddition.cpp
+++ b/tests/validation/reference/ArithmeticOperations.cpp
@@ -21,7 +21,7 @@
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
-#include "ArithmeticAddition.h"
+#include "ArithmeticOperations.h"
 
 #include "arm_compute/core/Types.h"
 #include "tests/validation/Helpers.h"
@@ -37,11 +37,12 @@ namespace reference
 namespace
 {
 template <typename T>
-T add(T src1, T src2, ConvertPolicy convert_policy)
+T arithm_op(ArithmeticOperation op, T src1, T src2, ConvertPolicy convert_policy)
 {
     using intermediate_type = typename common_promoted_signed_type<T>::intermediate_type;
 
-    intermediate_type val = static_cast<intermediate_type>(src1) + static_cast<intermediate_type>(src2);
+    intermediate_type val = (op == ArithmeticOperation::ADD) ? static_cast<intermediate_type>(src1) + static_cast<intermediate_type>(src2) : static_cast<intermediate_type>
+                            (src1) - static_cast<intermediate_type>(src2);
 
     T result = (convert_policy == ConvertPolicy::SATURATE) ? saturate_cast<T>(val) : static_cast<T>(val);
 
@@ -52,7 +53,7 @@ template <size_t dim>
 struct BroadcastUnroll
 {
     template <typename T>
-    static void unroll(const SimpleTensor<T> &src1, const SimpleTensor<T> &src2, SimpleTensor<T> &dst,
+    static void unroll(ArithmeticOperation op, const SimpleTensor<T> &src1, const SimpleTensor<T> &src2, SimpleTensor<T> &dst,
                        ConvertPolicy convert_policy, Coordinates &id_src1, Coordinates &id_src2, Coordinates &id_dst)
     {
         const bool src1_is_broadcast = (src1.shape()[dim - 1] != dst.shape()[dim - 1]);
@@ -64,7 +65,7 @@ struct BroadcastUnroll
 
         for(size_t i = 0; i < dst.shape()[dim - 1]; ++i, ++id_dst[dim - 1])
         {
-            BroadcastUnroll < dim - 1 >::unroll(src1, src2, dst, convert_policy, id_src1, id_src2, id_dst);
+            BroadcastUnroll < dim - 1 >::unroll(op, src1, src2, dst, convert_policy, id_src1, id_src2, id_dst);
 
             id_src1[dim - 1] += !src1_is_broadcast;
             id_src2[dim - 1] += !src2_is_broadcast;
@@ -76,26 +77,26 @@ template <>
 struct BroadcastUnroll<0>
 {
     template <typename T>
-    static void unroll(const SimpleTensor<T> &src1, const SimpleTensor<T> &src2, SimpleTensor<T> &dst,
+    static void unroll(ArithmeticOperation op, const SimpleTensor<T> &src1, const SimpleTensor<T> &src2, SimpleTensor<T> &dst,
                        ConvertPolicy convert_policy, Coordinates &id_src1, Coordinates &id_src2, Coordinates &id_dst)
     {
-        dst[coord2index(dst.shape(), id_dst)] = add(src1[coord2index(src1.shape(), id_src1)], src2[coord2index(src2.shape(), id_src2)], convert_policy);
+        dst[coord2index(dst.shape(), id_dst)] = arithm_op(op, src1[coord2index(src1.shape(), id_src1)], src2[coord2index(src2.shape(), id_src2)], convert_policy);
     }
 };
 } // namespace
 
 template <typename T>
-SimpleTensor<T> arithmetic_addition(const SimpleTensor<T> &src1, const SimpleTensor<T> &src2, SimpleTensor<T> &dst, ConvertPolicy convert_policy)
+SimpleTensor<T> arithmetic_operation(ArithmeticOperation op, const SimpleTensor<T> &src1, const SimpleTensor<T> &src2, SimpleTensor<T> &dst, ConvertPolicy convert_policy)
 {
     Coordinates id_src1, id_src2, id_dst;
 
-    BroadcastUnroll<Coordinates::num_max_dimensions>::unroll(src1, src2, dst, convert_policy, id_src1, id_src2, id_dst);
+    BroadcastUnroll<Coordinates::num_max_dimensions>::unroll(op, src1, src2, dst, convert_policy, id_src1, id_src2, id_dst);
 
     return dst;
 }
 
 template <>
-SimpleTensor<uint8_t> arithmetic_addition(const SimpleTensor<uint8_t> &src1, const SimpleTensor<uint8_t> &src2, SimpleTensor<uint8_t> &dst, ConvertPolicy convert_policy)
+SimpleTensor<uint8_t> arithmetic_operation(ArithmeticOperation op, const SimpleTensor<uint8_t> &src1, const SimpleTensor<uint8_t> &src2, SimpleTensor<uint8_t> &dst, ConvertPolicy convert_policy)
 {
     if(dst.data_type() == DataType::QASYMM8)
     {
@@ -105,7 +106,7 @@ SimpleTensor<uint8_t> arithmetic_addition(const SimpleTensor<uint8_t> &src1, con
 
         Coordinates id_src1, id_src2, id_dst;
 
-        BroadcastUnroll<Coordinates::num_max_dimensions>::unroll(src1_tmp, src2_tmp, dst_tmp, convert_policy, id_src1, id_src2, id_dst);
+        BroadcastUnroll<Coordinates::num_max_dimensions>::unroll(op, src1_tmp, src2_tmp, dst_tmp, convert_policy, id_src1, id_src2, id_dst);
 
         dst = convert_to_asymmetric(dst_tmp, dst.quantization_info());
         return dst;
@@ -115,31 +116,32 @@ SimpleTensor<uint8_t> arithmetic_addition(const SimpleTensor<uint8_t> &src1, con
         // DataType::U8
         Coordinates id_src1, id_src2, id_dst;
 
-        BroadcastUnroll<Coordinates::num_max_dimensions>::unroll(src1, src2, dst, convert_policy, id_src1, id_src2, id_dst);
+        BroadcastUnroll<Coordinates::num_max_dimensions>::unroll(op, src1, src2, dst, convert_policy, id_src1, id_src2, id_dst);
 
         return dst;
     }
 }
 
-template SimpleTensor<int16_t> arithmetic_addition(const SimpleTensor<int16_t> &src1, const SimpleTensor<int16_t> &src2, SimpleTensor<int16_t> &dst, ConvertPolicy convert_policy);
-template SimpleTensor<int8_t> arithmetic_addition(const SimpleTensor<int8_t> &src1, const SimpleTensor<int8_t> &src2, SimpleTensor<int8_t> &dst, ConvertPolicy convert_policy);
-template SimpleTensor<half> arithmetic_addition(const SimpleTensor<half> &src1, const SimpleTensor<half> &src2, SimpleTensor<half> &dst, ConvertPolicy convert_policy);
-template SimpleTensor<float> arithmetic_addition(const SimpleTensor<float> &src1, const SimpleTensor<float> &src2, SimpleTensor<float> &dst, ConvertPolicy convert_policy);
+template SimpleTensor<int16_t> arithmetic_operation(ArithmeticOperation op, const SimpleTensor<int16_t> &src1, const SimpleTensor<int16_t> &src2, SimpleTensor<int16_t> &dst,
+                                                    ConvertPolicy convert_policy);
+template SimpleTensor<int8_t> arithmetic_operation(ArithmeticOperation op, const SimpleTensor<int8_t> &src1, const SimpleTensor<int8_t> &src2, SimpleTensor<int8_t> &dst, ConvertPolicy convert_policy);
+template SimpleTensor<half> arithmetic_operation(ArithmeticOperation op, const SimpleTensor<half> &src1, const SimpleTensor<half> &src2, SimpleTensor<half> &dst, ConvertPolicy convert_policy);
+template SimpleTensor<float> arithmetic_operation(ArithmeticOperation op, const SimpleTensor<float> &src1, const SimpleTensor<float> &src2, SimpleTensor<float> &dst, ConvertPolicy convert_policy);
 
 template <typename T>
-SimpleTensor<T> arithmetic_addition(const SimpleTensor<T> &src1, const SimpleTensor<T> &src2, DataType dst_data_type, ConvertPolicy convert_policy)
+SimpleTensor<T> arithmetic_operation(ArithmeticOperation op, const SimpleTensor<T> &src1, const SimpleTensor<T> &src2, DataType dst_data_type, ConvertPolicy convert_policy)
 {
     ARM_COMPUTE_ERROR_ON_MSG(dst_data_type == DataType::QASYMM8, "For QASYMM8, the quantized output tensor should be passed directly.");
 
     SimpleTensor<T> dst(TensorShape::broadcast_shape(src1.shape(), src2.shape()), dst_data_type);
-    arithmetic_addition<T>(src1, src2, dst, convert_policy);
+    arithmetic_operation<T>(op, src1, src2, dst, convert_policy);
     return dst;
 }
 
-template SimpleTensor<int16_t> arithmetic_addition(const SimpleTensor<int16_t> &src1, const SimpleTensor<int16_t> &src2, DataType dst_data_type, ConvertPolicy convert_policy);
-template SimpleTensor<int8_t> arithmetic_addition(const SimpleTensor<int8_t> &src1, const SimpleTensor<int8_t> &src2, DataType dst_data_type, ConvertPolicy convert_policy);
-template SimpleTensor<half> arithmetic_addition(const SimpleTensor<half> &src1, const SimpleTensor<half> &src2, DataType dst_data_type, ConvertPolicy convert_policy);
-template SimpleTensor<float> arithmetic_addition(const SimpleTensor<float> &src1, const SimpleTensor<float> &src2, DataType dst_data_type, ConvertPolicy convert_policy);
+template SimpleTensor<int16_t> arithmetic_operation(ArithmeticOperation op, const SimpleTensor<int16_t> &src1, const SimpleTensor<int16_t> &src2, DataType dst_data_type, ConvertPolicy convert_policy);
+template SimpleTensor<int8_t> arithmetic_operation(ArithmeticOperation op, const SimpleTensor<int8_t> &src1, const SimpleTensor<int8_t> &src2, DataType dst_data_type, ConvertPolicy convert_policy);
+template SimpleTensor<half> arithmetic_operation(ArithmeticOperation op, const SimpleTensor<half> &src1, const SimpleTensor<half> &src2, DataType dst_data_type, ConvertPolicy convert_policy);
+template SimpleTensor<float> arithmetic_operation(ArithmeticOperation op, const SimpleTensor<float> &src1, const SimpleTensor<float> &src2, DataType dst_data_type, ConvertPolicy convert_policy);
 
 } // namespace reference
 } // namespace validation
diff --git a/tests/validation/reference/ArithmeticAddition.h b/tests/validation/reference/ArithmeticOperations.h
index faeabd7a6f..73634823c4 100644
--- a/tests/validation/reference/ArithmeticAddition.h
+++ b/tests/validation/reference/ArithmeticOperations.h
@@ -21,8 +21,8 @@
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
-#ifndef __ARM_COMPUTE_TEST_ARITHMETIC_ADDITION_H__
-#define __ARM_COMPUTE_TEST_ARITHMETIC_ADDITION_H__
+#ifndef __ARM_COMPUTE_TEST_ARITHMETIC_OPERATIONS_H__
+#define __ARM_COMPUTE_TEST_ARITHMETIC_OPERATIONS_H__
 
 #include "tests/SimpleTensor.h"
 #include "tests/validation/Helpers.h"
@@ -35,14 +35,20 @@ namespace validation
 {
 namespace reference
 {
-template <typename T>
-SimpleTensor<T> arithmetic_addition(const SimpleTensor<T> &src1, const SimpleTensor<T> &src2, SimpleTensor<T> &dst, ConvertPolicy convert_policy);
+/** Arithmetic operation types */
+enum class ArithmeticOperation
+{
+    ADD,
+    SUB
+};
 
 template <typename T>
-SimpleTensor<T> arithmetic_addition(const SimpleTensor<T> &src1, const SimpleTensor<T> &src2, DataType dst_data_type, ConvertPolicy convert_policy);
+SimpleTensor<T> arithmetic_operation(ArithmeticOperation op, const SimpleTensor<T> &src1, const SimpleTensor<T> &src2, SimpleTensor<T> &dst, ConvertPolicy convert_policy);
 
+template <typename T>
+SimpleTensor<T> arithmetic_operation(ArithmeticOperation op, const SimpleTensor<T> &src1, const SimpleTensor<T> &src2, DataType dst_data_type, ConvertPolicy convert_policy);
 } // namespace reference
 } // namespace validation
 } // namespace test
 } // namespace arm_compute
-#endif /* __ARM_COMPUTE_TEST_ARITHMETIC_ADDITION_H__ */
+#endif /* __ARM_COMPUTE_TEST_ARITHMETIC_OPERATIONS_H__ */
diff --git a/tests/validation/reference/ArithmeticSubtraction.cpp b/tests/validation/reference/ArithmeticSubtraction.cpp
deleted file mode 100644
index f39d01f9e8..0000000000
--- a/tests/validation/reference/ArithmeticSubtraction.cpp
+++ /dev/null
@@ -1,63 +0,0 @@
-/*
- * Copyright (c) 2017-2018 ARM Limited.
- *
- * SPDX-License-Identifier: MIT
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to
- * deal in the Software without restriction, including without limitation the
- * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
- * sell copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in all
- * copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-#include "ArithmeticSubtraction.h"
-
-#include "tests/validation/Helpers.h"
-
-namespace arm_compute
-{
-namespace test
-{
-namespace validation
-{
-namespace reference
-{
-template <typename T1, typename T2, typename T3>
-SimpleTensor<T3> arithmetic_subtraction(const SimpleTensor<T1> &src1, const SimpleTensor<T2> &src2, DataType dst_data_type, ConvertPolicy convert_policy)
-{
-    SimpleTensor<T3> result(src1.shape(), dst_data_type);
-
-    using intermediate_type = typename common_promoted_signed_type<typename std::conditional<sizeof(T1) >= sizeof(T2), T1, T2>::type >::intermediate_type;
-
-    for(int i = 0; i < src1.num_elements(); ++i)
-    {
-        intermediate_type val = static_cast<intermediate_type>(src1[i]) - static_cast<intermediate_type>(src2[i]);
-        result[i]             = (convert_policy == ConvertPolicy::SATURATE) ? saturate_cast<T3>(val) : static_cast<T3>(val);
-    }
-
-    return result;
-}
-
-template SimpleTensor<uint8_t> arithmetic_subtraction(const SimpleTensor<uint8_t> &src1, const SimpleTensor<uint8_t> &src2, DataType dst_data_type, ConvertPolicy convert_policy);
-template SimpleTensor<int16_t> arithmetic_subtraction(const SimpleTensor<uint8_t> &src1, const SimpleTensor<uint8_t> &src2, DataType dst_data_type, ConvertPolicy convert_policy);
-template SimpleTensor<int16_t> arithmetic_subtraction(const SimpleTensor<uint8_t> &src1, const SimpleTensor<int16_t> &src2, DataType dst_data_type, ConvertPolicy convert_policy);
-template SimpleTensor<int16_t> arithmetic_subtraction(const SimpleTensor<int16_t> &src1, const SimpleTensor<uint8_t> &src2, DataType dst_data_type, ConvertPolicy convert_policy);
-template SimpleTensor<int16_t> arithmetic_subtraction(const SimpleTensor<int16_t> &src1, const SimpleTensor<int16_t> &src2, DataType dst_data_type, ConvertPolicy convert_policy);
-template SimpleTensor<int8_t> arithmetic_subtraction(const SimpleTensor<int8_t> &src1, const SimpleTensor<int8_t> &src2, DataType dst_data_type, ConvertPolicy convert_policy);
-template SimpleTensor<half> arithmetic_subtraction(const SimpleTensor<half> &src1, const SimpleTensor<half> &src2, DataType dst_data_type, ConvertPolicy convert_policy);
-template SimpleTensor<float> arithmetic_subtraction(const SimpleTensor<float> &src1, const SimpleTensor<float> &src2, DataType dst_data_type, ConvertPolicy convert_policy);
-} // namespace reference
-} // namespace validation
-} // namespace test
-} // namespace arm_compute
diff --git a/tests/validation/reference/ArithmeticSubtraction.h b/tests/validation/reference/ArithmeticSubtraction.h
deleted file mode 100644
index 9308314bda..0000000000
--- a/tests/validation/reference/ArithmeticSubtraction.h
+++ /dev/null
@@ -1,44 +0,0 @@
-/*
- * Copyright (c) 2017 ARM Limited.
- *
- * SPDX-License-Identifier: MIT
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to
- * deal in the Software without restriction, including without limitation the
- * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
- * sell copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in all
- * copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-#ifndef __ARM_COMPUTE_TEST_ARITHMETIC_SUBTRACTION_H__
-#define __ARM_COMPUTE_TEST_ARITHMETIC_SUBTRACTION_H__
-
-#include "tests/SimpleTensor.h"
-#include "tests/validation/Helpers.h"
-
-namespace arm_compute
-{
-namespace test
-{
-namespace validation
-{
-namespace reference
-{
-template <typename T1, typename T2, typename T3>
-SimpleTensor<T3> arithmetic_subtraction(const SimpleTensor<T1> &src1, const SimpleTensor<T2> &src2, DataType dst_data_type, ConvertPolicy convert_policy);
-} // namespace reference
-} // namespace validation
-} // namespace test
-} // namespace arm_compute
-#endif /* __ARM_COMPUTE_TEST_ARITHMETIC_SUBTRACTION_H__ */
diff --git a/tests/validation/reference/LaplacianPyramid.cpp b/tests/validation/reference/LaplacianPyramid.cpp
index 566847453b..21ddc1e48e 100644
--- a/tests/validation/reference/LaplacianPyramid.cpp
+++ b/tests/validation/reference/LaplacianPyramid.cpp
@@ -23,7 +23,7 @@
  */
 #include "LaplacianPyramid.h"
 
-#include "tests/validation/reference/ArithmeticSubtraction.h"
+#include "tests/validation/reference/ArithmeticOperations.h"
 #include "tests/validation/reference/DepthConvertLayer.h"
 #include "tests/validation/reference/Gaussian5x5.h"
 #include "tests/validation/reference/GaussianPyramidHalf.h"
@@ -53,7 +53,10 @@ std::vector<SimpleTensor<U>> laplacian_pyramid(const SimpleTensor<T> &src, Simpl
         const SimpleTensor<T> level_filtered = reference::gaussian5x5(gaussian_level_pyramid[i], border_mode, constant_border_value);
         pyramid_conv.push_back(level_filtered);
 
-        const SimpleTensor<U> level_sub = reference::arithmetic_subtraction<T, T, U>(gaussian_level_pyramid[i], level_filtered, dst.data_type(), ConvertPolicy::WRAP);
+        const SimpleTensor<U> level_filtered_converted = depth_convert<T, U>(level_filtered, DataType::S16, ConvertPolicy::WRAP, 0);
+        const SimpleTensor<U> gaussian_level_converted = depth_convert<T, U>(gaussian_level_pyramid[i], DataType::S16, ConvertPolicy::WRAP, 0);
+
+        const SimpleTensor<U> level_sub = reference::arithmetic_operation<U>(reference::ArithmeticOperation::SUB, gaussian_level_converted, level_filtered_converted, dst.data_type(), ConvertPolicy::WRAP);
         pyramid_dst.push_back(level_sub);
     }
 
diff --git a/tests/validation/reference/LaplacianReconstruct.cpp b/tests/validation/reference/LaplacianReconstruct.cpp
index 23468287b0..ef14355bc8 100644
--- a/tests/validation/reference/LaplacianReconstruct.cpp
+++ b/tests/validation/reference/LaplacianReconstruct.cpp
@@ -24,7 +24,7 @@
 #include "LaplacianReconstruct.h"
 
 #include "arm_compute/core/Types.h"
-#include "tests/validation/reference/ArithmeticAddition.h"
+#include "tests/validation/reference/ArithmeticOperations.h"
 #include "tests/validation/reference/DepthConvertLayer.h"
 #include "tests/validation/reference/Scale.h"
 
@@ -45,7 +45,7 @@ SimpleTensor<U> laplacian_reconstruct(const std::vector<SimpleTensor<T>> &pyrami
     const DataType data_type  = low_res.data_type();
 
     // input + L(n-1)
-    tmp_pyramid[last_level] = reference::arithmetic_addition(low_res, pyramid[last_level], data_type, ConvertPolicy::SATURATE);
+    tmp_pyramid[last_level] = reference::arithmetic_operation(reference::ArithmeticOperation::ADD, low_res, pyramid[last_level], data_type, ConvertPolicy::SATURATE);
 
     // Scale levels n-1 to 1, and add levels n-2 to 0
     for(size_t i = last_level; i-- > 0;)
@@ -56,7 +56,7 @@ SimpleTensor<U> laplacian_reconstruct(const std::vector<SimpleTensor<T>> &pyrami
         tmp_pyramid[i] = reference::scale(tmp_pyramid[i + 1], scale_x, scale_y, InterpolationPolicy::NEAREST_NEIGHBOR,
                                           border_mode, constant_border_value, SamplingPolicy::CENTER, false);
 
-        tmp_pyramid[i] = reference::arithmetic_addition(tmp_pyramid[i], pyramid[i], data_type, ConvertPolicy::SATURATE);
+        tmp_pyramid[i] = reference::arithmetic_operation(reference::ArithmeticOperation::ADD, tmp_pyramid[i], pyramid[i], data_type, ConvertPolicy::SATURATE);
     }
 
     return reference::depth_convert<T, U>(tmp_pyramid[0], DataType::U8, ConvertPolicy::SATURATE, 0);