Port CLCol2ImKernel to ClCol2ImKernel

Resolves: COMPMID-4517

Change-Id: I50cb02116a1ab86fc29200371944c4774e830746
Signed-off-by: Manuel Bottini <manuel.bottini@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5949
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>

3 files changed, 51 insertions, 13 deletions

diff --git a/tests/CL/Helper.h b/tests/CL/Helper.h
index b99911e1e6..e3e64c9405 100644
--- a/tests/CL/Helper.h
+++ b/tests/CL/Helper.h
@@ -218,6 +218,39 @@ private:
     CLFillBorderKernel         _border_handler{}; /**< Kernel to handle  borders */
     std::unique_ptr<ICLKernel> _kernel{};         /**< Kernel to run */
 };
+
+/** As above but this also setups a Zero border on the input tensor of the kernel's bordersize */
+template <typename K>
+class ClSynthetizeOperatorWithBorder : public opencl::IClOperator
+{
+public:
+    /** Configure the kernel.
+     *
+     * @param[in] first First configuration argument.
+     * @param[in] args  Rest of the configuration arguments.
+     */
+    template <typename T, typename... Args>
+    void configure(T first, Args &&... args)
+    {
+        auto k = std::make_unique<K>();
+        k->configure(CLKernelLibrary::get().get_compile_context(), first, std::forward<Args>(args)...);
+        _kernel = std::move(k);
+
+        auto b = std::make_unique<CLFillBorderKernel>();
+        b->configure(CLKernelLibrary::get().get_compile_context(), first, BorderSize(_kernel->border_size()), BorderMode::CONSTANT, PixelValue());
+        _border_handler = std::move(b);
+    }
+
+    void run(ITensorPack &tensors) override
+    {
+        CLScheduler::get().enqueue(*_border_handler);
+        CLScheduler::get().enqueue_op(*_kernel, tensors);
+    }
+
+private:
+    std::unique_ptr<ICLKernel> _border_handler{ nullptr }; /**< Kernel to handle  borders */
+    std::unique_ptr<ICLKernel> _kernel{};                  /**< Kernel to run */
+};
 } // namespace test
 } // namespace arm_compute
 #endif /* ARM_COMPUTE_TEST_CL_HELPER_H */
diff --git a/tests/validation/CL/Col2Im.cpp b/tests/validation/CL/Col2Im.cpp
index b651bf8918..96096d314d 100644
--- a/tests/validation/CL/Col2Im.cpp
+++ b/tests/validation/CL/Col2Im.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018-2020 Arm Limited.
+ * Copyright (c) 2018-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -22,7 +22,7 @@
  * SOFTWARE.
  */
 #include "arm_compute/core/Types.h"
-#include "src/core/CL/kernels/CLCol2ImKernel.h"
+#include "src/core/gpu/cl/kernels/ClCol2ImKernel.h"
 #include "tests/CL/CLAccessor.h"
 #include "tests/CL/Helper.h"
 #include "tests/framework/Asserts.h"
@@ -40,7 +40,7 @@ namespace validation
 TEST_SUITE(CL)
 TEST_SUITE(Col2Im)
 
-using CLCol2Im = CLSynthetizeFunction<CLCol2ImKernel>;
+using ClCol2Im = ClSynthetizeOperatorWithBorder<opencl::kernels::ClCol2ImKernel>;
 
 /** Negative tests
  *
@@ -59,7 +59,7 @@ TEST_CASE(Negative, framework::DatasetMode::ALL)
         const auto input     = TensorInfo(TensorShape(10U, 12U, 1U, 2U), 1, DataType::SIZET);
         const auto output    = TensorInfo(TensorShape(3U, 4U, 10U, 1U, 2U), 1, DataType::F32);
         const auto conv_size = Size2D(3, 4);
-        const auto status    = CLCol2ImKernel::validate(&input, &output, conv_size);
+        const auto status    = opencl::kernels::ClCol2ImKernel::validate(&input, &output, conv_size);
         ARM_COMPUTE_EXPECT(bool(status) == false, framework::LogLevel::ERRORS);
     }
 
@@ -68,7 +68,7 @@ TEST_CASE(Negative, framework::DatasetMode::ALL)
         const auto input     = TensorInfo(TensorShape(10U, 12U, 1U, 2U), 1, DataType::F32);
         const auto output    = TensorInfo(TensorShape(3U, 4U, 10U, 1U, 2U), 1, DataType::F32, DataLayout::NHWC);
         const auto conv_size = Size2D(3, 4);
-        const auto status    = CLCol2ImKernel::validate(&input, &output, conv_size);
+        const auto status    = opencl::kernels::ClCol2ImKernel::validate(&input, &output, conv_size);
         ARM_COMPUTE_EXPECT(bool(status) == false, framework::LogLevel::ERRORS);
     }
 
@@ -77,13 +77,13 @@ TEST_CASE(Negative, framework::DatasetMode::ALL)
         const auto input     = TensorInfo(TensorShape(10U, 12U, 1U, 2U), 1, DataType::F32);
         const auto output    = TensorInfo(TensorShape(3U, 4U, 10U, 2U, 2U), 1, DataType::F32);
         const auto conv_size = Size2D(3, 4);
-        const auto status    = CLCol2ImKernel::validate(&input, &output, conv_size);
+        const auto status    = opencl::kernels::ClCol2ImKernel::validate(&input, &output, conv_size);
         ARM_COMPUTE_EXPECT(bool(status) == false, framework::LogLevel::ERRORS);
     }
 }
 
 template <typename T>
-using CLCol2ImFixture = Col2ImValidationFixture<CLTensor, CLAccessor, CLCol2Im, T, true>;
+using ClCol2ImFixture = Col2ImOpValidationFixture<CLTensor, CLAccessor, ClCol2Im, T, true>;
 
 /** Test kernel for single-precision floating point
  *
@@ -99,7 +99,7 @@ using CLCol2ImFixture = Col2ImValidationFixture<CLTensor, CLAccessor, CLCol2Im,
  *  Kernel tested col2im
  */
 FIXTURE_DATA_TEST_CASE(FP32,
-                       CLCol2ImFixture<float>,
+                       ClCol2ImFixture<float>,
                        framework::DatasetMode::ALL,
                        combine(combine(combine(combine(
                                                    framework::dataset::make("InputShape", { TensorShape(8U, 16U, 3U, 1U), TensorShape(17U, 16U, 3U, 1U), TensorShape(7U, 16U, 3U, 1U) }),
@@ -125,7 +125,7 @@ FIXTURE_DATA_TEST_CASE(FP32,
  *  Kernel tested col2im
  */
 FIXTURE_DATA_TEST_CASE(F16,
-                       CLCol2ImFixture<half>,
+                       ClCol2ImFixture<half>,
                        framework::DatasetMode::ALL,
                        combine(combine(combine(combine(
                                                    framework::dataset::make("InputShape", TensorShape(17U, 16U, 3U, 1U)),
@@ -151,7 +151,7 @@ FIXTURE_DATA_TEST_CASE(F16,
  *  Kernel tested col2im
  */
 FIXTURE_DATA_TEST_CASE(QASYMM8,
-                       CLCol2ImFixture<uint8_t>,
+                       ClCol2ImFixture<uint8_t>,
                        framework::DatasetMode::ALL,
                        combine(combine(combine(combine(
                                                    framework::dataset::make("InputShape", TensorShape(17U, 16U, 3U, 1U)),
diff --git a/tests/validation/fixtures/Col2ImFixture.h b/tests/validation/fixtures/Col2ImFixture.h
index 88d420aa3d..ee7a14f844 100644
--- a/tests/validation/fixtures/Col2ImFixture.h
+++ b/tests/validation/fixtures/Col2ImFixture.h
@@ -45,7 +45,7 @@ namespace validation
 using namespace arm_compute::misc::shape_calculator;
 
 template <typename TensorType, typename AccessorType, typename FunctionType, typename T, bool batch_size_on_z>
-class Col2ImValidationFixture : public framework::Fixture
+class Col2ImOpValidationFixture : public framework::Fixture
 {
 public:
     template <typename...>
@@ -74,7 +74,7 @@ protected:
 
         // Create and configure function
         FunctionType col2im_func;
-        col2im_func.configure(&src, &dst, convolved_dims, num_groups);
+        col2im_func.configure(src.info(), dst.info(), convolved_dims, num_groups);
 
         ARM_COMPUTE_ASSERT(src.info()->is_resizable());
         ARM_COMPUTE_ASSERT(dst.info()->is_resizable());
@@ -89,8 +89,13 @@ protected:
         // Fill tensors
         fill(AccessorType(src), 0);
 
+        arm_compute::ITensorPack pack =
+        {
+            { arm_compute::TensorType::ACL_SRC, &src },
+            { arm_compute::TensorType::ACL_DST, &dst }
+        };
         // Compute function
-        col2im_func.run();
+        col2im_func.run(pack);
 
         return dst;
     }