COMPMID-1718: Extend DepthConvert to support Cast

Change-Id: I6ee2c0b670727fc808fa636c53ddfaec3a0036c9

5 files changed, 467 insertions, 102 deletions

diff --git a/tests/validation/CL/Cast.cpp b/tests/validation/CL/Cast.cpp
new file mode 100644
index 0000000000..107edcd11e
--- /dev/null
+++ b/tests/validation/CL/Cast.cpp
@@ -0,0 +1,244 @@
+/*
+ * Copyright (c) 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 CONCLCTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "arm_compute/core/Types.h"
+#include "arm_compute/runtime/CL/CLTensor.h"
+#include "arm_compute/runtime/CL/CLTensorAllocator.h"
+#include "arm_compute/runtime/CL/functions/CLCast.h"
+#include "tests/CL/CLAccessor.h"
+#include "tests/PaddingCalculator.h"
+#include "tests/datasets/ConvertPolicyDataset.h"
+#include "tests/datasets/ShapeDatasets.h"
+#include "tests/framework/Asserts.h"
+#include "tests/framework/Macros.h"
+#include "tests/framework/datasets/Datasets.h"
+#include "tests/validation/Validation.h"
+#include "tests/validation/fixtures/CastFixture.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace
+{
+// Tolerance
+constexpr AbsoluteTolerance<float> one_tolerance(1);
+constexpr AbsoluteTolerance<float> zero_tolerance(0);
+
+/** Input data sets **/
+// U8
+const auto CastU8toS8Dataset  = combine(framework::dataset::make("DataType", DataType::U8), framework::dataset::make("DataType", DataType::S8));
+const auto CastU8toU16Dataset = combine(framework::dataset::make("DataType", DataType::U8), framework::dataset::make("DataType", DataType::U16));
+const auto CastU8toS16Dataset = combine(framework::dataset::make("DataType", DataType::U8), framework::dataset::make("DataType", DataType::S16));
+const auto CastU8toU32Dataset = combine(framework::dataset::make("DataType", DataType::U8), framework::dataset::make("DataType", DataType::U32));
+const auto CastU8toS32Dataset = combine(framework::dataset::make("DataType", DataType::U8), framework::dataset::make("DataType", DataType::S32));
+const auto CastU8toF16Dataset = combine(framework::dataset::make("DataType", DataType::U8), framework::dataset::make("DataType", DataType::F16));
+const auto CastU8toF32Dataset = combine(framework::dataset::make("DataType", DataType::U8), framework::dataset::make("DataType", DataType::F32));
+
+// S8
+const auto CastS8toU8Dataset  = combine(framework::dataset::make("DataType", DataType::S8), framework::dataset::make("DataType", DataType::U8));
+const auto CastS8toU16Dataset = combine(framework::dataset::make("DataType", DataType::S8), framework::dataset::make("DataType", DataType::U16));
+const auto CastS8toS16Dataset = combine(framework::dataset::make("DataType", DataType::S8), framework::dataset::make("DataType", DataType::S16));
+const auto CastS8toU32Dataset = combine(framework::dataset::make("DataType", DataType::S8), framework::dataset::make("DataType", DataType::U32));
+const auto CastS8toS32Dataset = combine(framework::dataset::make("DataType", DataType::S8), framework::dataset::make("DataType", DataType::S32));
+const auto CastS8toF16Dataset = combine(framework::dataset::make("DataType", DataType::S8), framework::dataset::make("DataType", DataType::F16));
+const auto CastS8toF32Dataset = combine(framework::dataset::make("DataType", DataType::S8), framework::dataset::make("DataType", DataType::F32));
+
+// U16
+const auto CastU16toU8Dataset  = combine(framework::dataset::make("DataType", DataType::U16), framework::dataset::make("DataType", DataType::U8));
+const auto CastU16toS8Dataset  = combine(framework::dataset::make("DataType", DataType::U16), framework::dataset::make("DataType", DataType::S8));
+const auto CastU16toS16Dataset = combine(framework::dataset::make("DataType", DataType::U16), framework::dataset::make("DataType", DataType::S16));
+const auto CastU16toU32Dataset = combine(framework::dataset::make("DataType", DataType::U16), framework::dataset::make("DataType", DataType::U32));
+const auto CastU16toS32Dataset = combine(framework::dataset::make("DataType", DataType::U16), framework::dataset::make("DataType", DataType::S32));
+const auto CastU16toF16Dataset = combine(framework::dataset::make("DataType", DataType::U16), framework::dataset::make("DataType", DataType::F16));
+const auto CastU16toF32Dataset = combine(framework::dataset::make("DataType", DataType::U16), framework::dataset::make("DataType", DataType::F32));
+
+// S16
+const auto CastS16toU8Dataset  = combine(framework::dataset::make("DataType", DataType::S16), framework::dataset::make("DataType", DataType::U8));
+const auto CastS16toS8Dataset  = combine(framework::dataset::make("DataType", DataType::S16), framework::dataset::make("DataType", DataType::S8));
+const auto CastS16toU16Dataset = combine(framework::dataset::make("DataType", DataType::S16), framework::dataset::make("DataType", DataType::U16));
+const auto CastS16toU32Dataset = combine(framework::dataset::make("DataType", DataType::S16), framework::dataset::make("DataType", DataType::U32));
+const auto CastS16toS32Dataset = combine(framework::dataset::make("DataType", DataType::S16), framework::dataset::make("DataType", DataType::S32));
+const auto CastS16toF16Dataset = combine(framework::dataset::make("DataType", DataType::S16), framework::dataset::make("DataType", DataType::F16));
+const auto CastS16toF32Dataset = combine(framework::dataset::make("DataType", DataType::S16), framework::dataset::make("DataType", DataType::F32));
+
+// U32
+const auto CastU32toU8Dataset  = combine(framework::dataset::make("DataType", DataType::U32), framework::dataset::make("DataType", DataType::U8));
+const auto CastU32toS8Dataset  = combine(framework::dataset::make("DataType", DataType::U32), framework::dataset::make("DataType", DataType::S8));
+const auto CastU32toU16Dataset = combine(framework::dataset::make("DataType", DataType::U32), framework::dataset::make("DataType", DataType::U16));
+const auto CastU32toS16Dataset = combine(framework::dataset::make("DataType", DataType::U32), framework::dataset::make("DataType", DataType::S16));
+const auto CastU32toS32Dataset = combine(framework::dataset::make("DataType", DataType::U32), framework::dataset::make("DataType", DataType::S32));
+const auto CastU32toF16Dataset = combine(framework::dataset::make("DataType", DataType::U32), framework::dataset::make("DataType", DataType::F16));
+const auto CastU32toF32Dataset = combine(framework::dataset::make("DataType", DataType::U32), framework::dataset::make("DataType", DataType::F32));
+
+// S32
+const auto CastS32toU8Dataset  = combine(framework::dataset::make("DataType", DataType::S32), framework::dataset::make("DataType", DataType::U8));
+const auto CastS32toS8Dataset  = combine(framework::dataset::make("DataType", DataType::S32), framework::dataset::make("DataType", DataType::S8));
+const auto CastS32toU16Dataset = combine(framework::dataset::make("DataType", DataType::S32), framework::dataset::make("DataType", DataType::U16));
+const auto CastS32toS16Dataset = combine(framework::dataset::make("DataType", DataType::S32), framework::dataset::make("DataType", DataType::S16));
+const auto CastS32toU32Dataset = combine(framework::dataset::make("DataType", DataType::S32), framework::dataset::make("DataType", DataType::U32));
+const auto CastS32toF16Dataset = combine(framework::dataset::make("DataType", DataType::S32), framework::dataset::make("DataType", DataType::F16));
+const auto CastS32toF32Dataset = combine(framework::dataset::make("DataType", DataType::S32), framework::dataset::make("DataType", DataType::F32));
+
+// F16
+const auto CastF16toU8Dataset  = combine(framework::dataset::make("DataType", DataType::F16), framework::dataset::make("DataType", DataType::U8));
+const auto CastF16toS8Dataset  = combine(framework::dataset::make("DataType", DataType::F16), framework::dataset::make("DataType", DataType::S8));
+const auto CastF16toU16Dataset = combine(framework::dataset::make("DataType", DataType::F16), framework::dataset::make("DataType", DataType::U16));
+const auto CastF16toS16Dataset = combine(framework::dataset::make("DataType", DataType::F16), framework::dataset::make("DataType", DataType::S16));
+const auto CastF16toU32Dataset = combine(framework::dataset::make("DataType", DataType::F16), framework::dataset::make("DataType", DataType::U32));
+const auto CastF16toS32Dataset = combine(framework::dataset::make("DataType", DataType::F16), framework::dataset::make("DataType", DataType::S32));
+const auto CastF16toF32Dataset = combine(framework::dataset::make("DataType", DataType::F16), framework::dataset::make("DataType", DataType::F32));
+
+// F32
+const auto CastF32toU8Dataset  = combine(framework::dataset::make("DataType", DataType::F32), framework::dataset::make("DataType", DataType::U8));
+const auto CastF32toS8Dataset  = combine(framework::dataset::make("DataType", DataType::F32), framework::dataset::make("DataType", DataType::S8));
+const auto CastF32toU16Dataset = combine(framework::dataset::make("DataType", DataType::F32), framework::dataset::make("DataType", DataType::U16));
+const auto CastF32toS16Dataset = combine(framework::dataset::make("DataType", DataType::F32), framework::dataset::make("DataType", DataType::S16));
+const auto CastF32toU32Dataset = combine(framework::dataset::make("DataType", DataType::F32), framework::dataset::make("DataType", DataType::U32));
+const auto CastF32toS32Dataset = combine(framework::dataset::make("DataType", DataType::F32), framework::dataset::make("DataType", DataType::S32));
+const auto CastF32toF16Dataset = combine(framework::dataset::make("DataType", DataType::F32), framework::dataset::make("DataType", DataType::F16));
+} // namespace
+
+TEST_SUITE(CL)
+TEST_SUITE(Cast)
+template <typename T>
+using CLCastToU8Fixture = CastValidationFixture<CLTensor, CLAccessor, CLCast, T, uint8_t>;
+template <typename T>
+using CLCastToS8Fixture = CastValidationFixture<CLTensor, CLAccessor, CLCast, T, int8_t>;
+template <typename T>
+using CLCastToU16Fixture = CastValidationFixture<CLTensor, CLAccessor, CLCast, T, uint16_t>;
+template <typename T>
+using CLCastToS16Fixture = CastValidationFixture<CLTensor, CLAccessor, CLCast, T, int16_t>;
+template <typename T>
+using CLCastToU32Fixture = CastValidationFixture<CLTensor, CLAccessor, CLCast, T, uint32_t>;
+template <typename T>
+using CLCastToS32Fixture = CastValidationFixture<CLTensor, CLAccessor, CLCast, T, int32_t>;
+template <typename T>
+using CLCastToF16Fixture = CastValidationFixture<CLTensor, CLAccessor, CLCast, T, half>;
+template <typename T>
+using CLCastToF32Fixture = CastValidationFixture<CLTensor, CLAccessor, CLCast, T, float>;
+
+#define CAST_SUITE(NAME, idt, odt, type, dataset, tolerance)                                                                     \
+    TEST_SUITE(NAME)                                                                                                             \
+    DATA_TEST_CASE(Configuration, framework::DatasetMode::ALL, combine(datasets::SmallShapes(), datasets::ConvertPolicies()),    \
+                   shape, policy)                                                                                                \
+    {                                                                                                                            \
+        CLTensor src = create_tensor<CLTensor>(shape, idt, 1);                                                                   \
+        CLTensor dst = create_tensor<CLTensor>(shape, odt, 1);                                                                   \
+        \
+        CLCast cast;                                                                                                             \
+        cast.configure(&src, &dst, policy);                                                                                      \
+        \
+        const ValidRegion valid_region = shape_to_valid_region(shape);                                                           \
+        validate(dst.info()->valid_region(), valid_region);                                                                      \
+        \
+        const PaddingSize padding = PaddingCalculator(shape.x(), 16).required_padding();                                         \
+        validate(src.info()->padding(), padding);                                                                                \
+        validate(dst.info()->padding(), padding);                                                                                \
+    }                                                                                                                            \
+    FIXTURE_DATA_TEST_CASE(RunSmall, type, framework::DatasetMode::PRECOMMIT, combine(combine(datasets::SmallShapes(), dataset), \
+                                                                                      datasets::ConvertPolicies()))              \
+    {                                                                                                                            \
+        validate(CLAccessor(_target), _reference, tolerance);                                                                    \
+    }                                                                                                                            \
+    TEST_SUITE_END()
+
+// U8
+CAST_SUITE(U8_to_S8, DataType::U8, DataType::S8, CLCastToS8Fixture<uint8_t>, CastU8toS8Dataset, zero_tolerance)
+CAST_SUITE(U8_to_U16, DataType::U8, DataType::U16, CLCastToU16Fixture<uint8_t>, CastU8toU16Dataset, zero_tolerance)
+CAST_SUITE(U8_to_S16, DataType::U8, DataType::S16, CLCastToS16Fixture<uint8_t>, CastU8toS16Dataset, zero_tolerance)
+CAST_SUITE(U8_to_U32, DataType::U8, DataType::U32, CLCastToU32Fixture<uint8_t>, CastU8toU32Dataset, zero_tolerance)
+CAST_SUITE(U8_to_S32, DataType::U8, DataType::S32, CLCastToS32Fixture<uint8_t>, CastU8toS32Dataset, zero_tolerance)
+CAST_SUITE(U8_to_F16, DataType::U8, DataType::F16, CLCastToF16Fixture<uint8_t>, CastU8toF16Dataset, zero_tolerance)
+CAST_SUITE(U8_to_F32, DataType::U8, DataType::F32, CLCastToF32Fixture<uint8_t>, CastU8toF32Dataset, zero_tolerance)
+
+// S8
+CAST_SUITE(S8_to_U8, DataType::S8, DataType::U8, CLCastToU8Fixture<int8_t>, CastS8toU8Dataset, zero_tolerance)
+CAST_SUITE(S8_to_U16, DataType::S8, DataType::U16, CLCastToU16Fixture<int8_t>, CastS8toU16Dataset, zero_tolerance)
+CAST_SUITE(S8_to_S16, DataType::S8, DataType::S16, CLCastToS16Fixture<int8_t>, CastS8toS16Dataset, zero_tolerance)
+CAST_SUITE(S8_to_U32, DataType::S8, DataType::U32, CLCastToU32Fixture<int8_t>, CastS8toU32Dataset, zero_tolerance)
+CAST_SUITE(S8_to_S32, DataType::S8, DataType::S32, CLCastToS32Fixture<int8_t>, CastS8toS32Dataset, zero_tolerance)
+CAST_SUITE(S8_to_F16, DataType::S8, DataType::F16, CLCastToF16Fixture<int8_t>, CastS8toF16Dataset, zero_tolerance)
+CAST_SUITE(S8_to_F32, DataType::S8, DataType::F32, CLCastToF32Fixture<int8_t>, CastS8toF32Dataset, zero_tolerance)
+
+// U16
+CAST_SUITE(U16_to_U8, DataType::U16, DataType::U8, CLCastToU8Fixture<uint16_t>, CastU16toU8Dataset, zero_tolerance)
+CAST_SUITE(U16_to_S8, DataType::U16, DataType::S8, CLCastToS8Fixture<uint16_t>, CastU16toS8Dataset, zero_tolerance)
+CAST_SUITE(U16_to_S16, DataType::U16, DataType::S16, CLCastToS16Fixture<uint16_t>, CastU16toS16Dataset, zero_tolerance)
+CAST_SUITE(U16_to_U32, DataType::U16, DataType::U32, CLCastToU32Fixture<uint16_t>, CastU16toU32Dataset, zero_tolerance)
+CAST_SUITE(U16_to_S32, DataType::U16, DataType::S32, CLCastToS32Fixture<uint16_t>, CastU16toS32Dataset, zero_tolerance)
+CAST_SUITE(U16_to_F16, DataType::U16, DataType::F16, CLCastToF16Fixture<uint16_t>, CastU16toF16Dataset, zero_tolerance)
+CAST_SUITE(U16_to_F32, DataType::U16, DataType::F32, CLCastToF32Fixture<uint16_t>, CastU16toF32Dataset, zero_tolerance)
+
+// S16
+CAST_SUITE(S16_to_U8, DataType::S16, DataType::U8, CLCastToU8Fixture<int16_t>, CastS16toU8Dataset, zero_tolerance)
+CAST_SUITE(S16_to_S8, DataType::S16, DataType::S8, CLCastToS8Fixture<int16_t>, CastS16toS8Dataset, zero_tolerance)
+CAST_SUITE(S16_to_U16, DataType::S16, DataType::U16, CLCastToU16Fixture<int16_t>, CastS16toU16Dataset, zero_tolerance)
+CAST_SUITE(S16_to_U32, DataType::S16, DataType::U32, CLCastToU32Fixture<int16_t>, CastS16toU32Dataset, zero_tolerance)
+CAST_SUITE(S16_to_S32, DataType::S16, DataType::S32, CLCastToS32Fixture<int16_t>, CastS16toS32Dataset, zero_tolerance)
+CAST_SUITE(S16_to_F16, DataType::S16, DataType::F16, CLCastToF16Fixture<int16_t>, CastS16toF16Dataset, zero_tolerance)
+CAST_SUITE(S16_to_F32, DataType::S16, DataType::F32, CLCastToF32Fixture<int16_t>, CastS16toF32Dataset, zero_tolerance)
+
+// U32
+CAST_SUITE(U32_to_U8, DataType::U32, DataType::U8, CLCastToU8Fixture<uint32_t>, CastU32toU8Dataset, zero_tolerance)
+CAST_SUITE(U32_to_S8, DataType::U32, DataType::S8, CLCastToS8Fixture<uint32_t>, CastU32toS8Dataset, zero_tolerance)
+CAST_SUITE(U32_to_U16, DataType::U32, DataType::U16, CLCastToU16Fixture<uint32_t>, CastU32toU16Dataset, zero_tolerance)
+CAST_SUITE(U32_to_S16, DataType::U32, DataType::S16, CLCastToS16Fixture<uint32_t>, CastU32toS16Dataset, zero_tolerance)
+CAST_SUITE(U32_to_S32, DataType::U32, DataType::S32, CLCastToS32Fixture<uint32_t>, CastU32toS32Dataset, zero_tolerance)
+CAST_SUITE(U32_to_F16, DataType::U32, DataType::F16, CLCastToF16Fixture<uint32_t>, CastU32toF16Dataset, zero_tolerance)
+CAST_SUITE(U32_to_F32, DataType::U32, DataType::F32, CLCastToF32Fixture<uint32_t>, CastU32toF32Dataset, zero_tolerance)
+
+// S32
+CAST_SUITE(S32_to_U8, DataType::S32, DataType::U8, CLCastToU8Fixture<int32_t>, CastS32toU8Dataset, zero_tolerance)
+CAST_SUITE(S32_to_S8, DataType::S32, DataType::S8, CLCastToS8Fixture<int32_t>, CastS32toS8Dataset, zero_tolerance)
+CAST_SUITE(S32_to_U16, DataType::S32, DataType::U16, CLCastToU16Fixture<int32_t>, CastS32toU16Dataset, zero_tolerance)
+CAST_SUITE(S32_to_S16, DataType::S32, DataType::S16, CLCastToS16Fixture<int32_t>, CastS32toS16Dataset, zero_tolerance)
+CAST_SUITE(S32_to_U32, DataType::S32, DataType::U32, CLCastToU32Fixture<int32_t>, CastS32toU32Dataset, zero_tolerance)
+CAST_SUITE(S32_to_F16, DataType::S32, DataType::F16, CLCastToF16Fixture<int32_t>, CastS32toF16Dataset, zero_tolerance)
+CAST_SUITE(S32_to_F32, DataType::S32, DataType::F32, CLCastToF32Fixture<int32_t>, CastS32toF32Dataset, zero_tolerance)
+
+// F16
+CAST_SUITE(F16_to_U8, DataType::F16, DataType::U8, CLCastToU8Fixture<half>, CastF16toU8Dataset, one_tolerance)
+CAST_SUITE(F16_to_S8, DataType::F16, DataType::S8, CLCastToS8Fixture<half>, CastF16toS8Dataset, one_tolerance)
+CAST_SUITE(F16_to_U16, DataType::F16, DataType::U16, CLCastToU16Fixture<half>, CastF16toU16Dataset, one_tolerance)
+CAST_SUITE(F16_to_S16, DataType::F16, DataType::S16, CLCastToS16Fixture<half>, CastF16toS16Dataset, one_tolerance)
+CAST_SUITE(F16_to_U32, DataType::F16, DataType::U32, CLCastToU32Fixture<half>, CastF16toU32Dataset, one_tolerance)
+CAST_SUITE(F16_to_S32, DataType::F16, DataType::S32, CLCastToS32Fixture<half>, CastF16toS32Dataset, one_tolerance)
+CAST_SUITE(F16_to_F32, DataType::F16, DataType::F32, CLCastToF32Fixture<half>, CastF16toF32Dataset, one_tolerance)
+
+// F32
+CAST_SUITE(F32_to_U8, DataType::F32, DataType::U8, CLCastToU8Fixture<float>, CastF32toU8Dataset, one_tolerance)
+CAST_SUITE(F32_to_S8, DataType::F32, DataType::S8, CLCastToS8Fixture<float>, CastF32toS8Dataset, one_tolerance)
+CAST_SUITE(F32_to_U16, DataType::F32, DataType::U16, CLCastToU16Fixture<float>, CastF32toU16Dataset, one_tolerance)
+CAST_SUITE(F32_to_S16, DataType::F32, DataType::S16, CLCastToS16Fixture<float>, CastF32toS16Dataset, one_tolerance)
+CAST_SUITE(F32_to_U32, DataType::F32, DataType::U32, CLCastToU32Fixture<float>, CastF32toU32Dataset, one_tolerance)
+CAST_SUITE(F32_to_S32, DataType::F32, DataType::S32, CLCastToS32Fixture<float>, CastF32toS32Dataset, one_tolerance)
+CAST_SUITE(F32_to_F16, DataType::F32, DataType::F16, CLCastToF16Fixture<float>, CastF32toF16Dataset, one_tolerance)
+
+TEST_SUITE_END() // Cast
+TEST_SUITE_END() // CL
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/CL/DepthConvertLayer.cpp b/tests/validation/CL/DepthConvertLayer.cpp
index fe46313568..7d7b5f2176 100644
--- a/tests/validation/CL/DepthConvertLayer.cpp
+++ b/tests/validation/CL/DepthConvertLayer.cpp
@@ -348,84 +348,6 @@ FIXTURE_DATA_TEST_CASE(RunLarge, CLDepthConvertLayerToS32Fixture<int16_t>, frame
 }
 TEST_SUITE_END()
 
-TEST_SUITE(F16_to_F32)
-DATA_TEST_CASE(Configuration, framework::DatasetMode::ALL, combine(combine(framework::dataset::concat(datasets::SmallShapes(), datasets::LargeShapes()), framework::dataset::make("ConvertPolicy", { ConvertPolicy::SATURATE, ConvertPolicy::WRAP })),
-                                                                   DepthConvertLayerShiftDataset),
-               shape, policy, shift)
-{
-    // Create tensors
-    CLTensor src = create_tensor<CLTensor>(shape, DataType::F16, 1);
-    CLTensor dst = create_tensor<CLTensor>(shape, DataType::F32, 1);
-
-    // Create and Configure function
-    CLDepthConvertLayer depth_convert;
-    depth_convert.configure(&src, &dst, policy, shift);
-
-    // Validate valid region
-    const ValidRegion valid_region = shape_to_valid_region(shape);
-    validate(dst.info()->valid_region(), valid_region);
-
-    // Validate padding
-    const PaddingSize padding = PaddingCalculator(shape.x(), 16).required_padding();
-    validate(src.info()->padding(), padding);
-    validate(dst.info()->padding(), padding);
-}
-
-FIXTURE_DATA_TEST_CASE(RunSmall, CLDepthConvertLayerToF32Fixture<half>, framework::DatasetMode::PRECOMMIT, combine(combine(combine(datasets::SmallShapes(), DepthConvertLayerF16toF32Dataset),
-                                                                                                                   framework::dataset::make("ConvertPolicy", { ConvertPolicy::SATURATE, ConvertPolicy::WRAP })),
-                                                                                                                   DepthConvertLayerShiftDataset))
-{
-    // Validate output
-    validate(CLAccessor(_target), _reference);
-}
-FIXTURE_DATA_TEST_CASE(RunLarge, CLDepthConvertLayerToF32Fixture<half>, framework::DatasetMode::NIGHTLY, combine(combine(combine(datasets::LargeShapes(), DepthConvertLayerF16toF32Dataset),
-                                                                                                                 framework::dataset::make("ConvertPolicy", { ConvertPolicy::SATURATE, ConvertPolicy::WRAP })),
-                                                                                                                 DepthConvertLayerShiftDataset))
-{
-    // Validate output
-    validate(CLAccessor(_target), _reference);
-}
-TEST_SUITE_END()
-
-TEST_SUITE(F32_to_F16)
-DATA_TEST_CASE(Configuration, framework::DatasetMode::ALL, combine(combine(framework::dataset::concat(datasets::SmallShapes(), datasets::LargeShapes()), framework::dataset::make("ConvertPolicy", { ConvertPolicy::SATURATE, ConvertPolicy::WRAP })),
-                                                                   DepthConvertLayerShiftDataset),
-               shape, policy, shift)
-{
-    // Create tensors
-    CLTensor src = create_tensor<CLTensor>(shape, DataType::F32, 1);
-    CLTensor dst = create_tensor<CLTensor>(shape, DataType::F16, 1);
-
-    // Create and Configure function
-    CLDepthConvertLayer depth_convert;
-    depth_convert.configure(&src, &dst, policy, shift);
-
-    // Validate valid region
-    const ValidRegion valid_region = shape_to_valid_region(shape);
-    validate(dst.info()->valid_region(), valid_region);
-
-    // Validate padding
-    const PaddingSize padding = PaddingCalculator(shape.x(), 16).required_padding();
-    validate(src.info()->padding(), padding);
-    validate(dst.info()->padding(), padding);
-}
-
-FIXTURE_DATA_TEST_CASE(RunSmall, CLDepthConvertLayerToF16Fixture<float>, framework::DatasetMode::PRECOMMIT, combine(combine(combine(datasets::SmallShapes(), DepthConvertLayerF32toF16Dataset),
-                                                                                                                    framework::dataset::make("ConvertPolicy", { ConvertPolicy::SATURATE, ConvertPolicy::WRAP })),
-                                                                                                                    DepthConvertLayerShiftDataset))
-{
-    // Validate output
-    validate(CLAccessor(_target), _reference);
-}
-FIXTURE_DATA_TEST_CASE(RunLarge, CLDepthConvertLayerToF16Fixture<float>, framework::DatasetMode::NIGHTLY, combine(combine(combine(datasets::LargeShapes(), DepthConvertLayerF32toF16Dataset),
-                                                                                                                  framework::dataset::make("ConvertPolicy", { ConvertPolicy::SATURATE, ConvertPolicy::WRAP })),
-                                                                                                                  DepthConvertLayerShiftDataset))
-{
-    // Validate output
-    validate(CLAccessor(_target), _reference);
-}
-TEST_SUITE_END()
-
 TEST_SUITE_END()
 TEST_SUITE_END()
 } // namespace validation
diff --git a/tests/validation/fixtures/CastFixture.h b/tests/validation/fixtures/CastFixture.h
new file mode 100644
index 0000000000..3a6efa22af
--- /dev/null
+++ b/tests/validation/fixtures/CastFixture.h
@@ -0,0 +1,144 @@
+/*
+ * Copyright (c) 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.
+ */
+#ifndef ARM_COMPUTE_TEST_CAST_FIXTURE
+#define ARM_COMPUTE_TEST_CAST_FIXTURE
+
+#include "tests/validation/fixtures/DepthConvertLayerFixture.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+template <typename TensorType, typename AccessorType, typename FunctionType, typename T1, typename T2>
+class CastValidationFixture : public framework::Fixture
+{
+public:
+    template <typename...>
+    void setup(TensorShape shape, DataType dt_in, DataType dt_out, ConvertPolicy policy)
+    {
+        _target    = compute_target(shape, dt_in, dt_out, policy);
+        _reference = compute_reference(shape, dt_in, dt_out, policy);
+    }
+
+protected:
+    template <typename U>
+    void fill(U &&tensor, int i, DataType dt_in, DataType dt_out)
+    {
+        // Restricting range to avoid inf values
+        if(dt_out == DataType::F16)
+        {
+            const int signed_min   = -32000;
+            const int signed_max   = 32000;
+            const int unsigned_min = 0;
+            const int unsigned_max = 65000;
+
+            switch(dt_in)
+            {
+                case DataType::U8:
+                case DataType::QASYMM8:
+                case DataType::S8:
+                case DataType::F32:
+                {
+                    library->fill_tensor_uniform(tensor, i);
+                    break;
+                }
+                case DataType::U16:
+                {
+                    library->fill_tensor_uniform(tensor, i, static_cast<uint16_t>(unsigned_min), static_cast<uint16_t>(unsigned_max));
+                    break;
+                }
+                case DataType::S16:
+                {
+                    library->fill_tensor_uniform(tensor, i, static_cast<int16_t>(signed_min), static_cast<int16_t>(signed_max));
+                    break;
+                }
+                case DataType::U32:
+                {
+                    library->fill_tensor_uniform(tensor, i, static_cast<uint32_t>(unsigned_min), static_cast<uint32_t>(unsigned_max));
+                    break;
+                }
+                case DataType::S32:
+                {
+                    library->fill_tensor_uniform(tensor, i, static_cast<int32_t>(signed_min), static_cast<int32_t>(signed_max));
+                    break;
+                }
+                default:
+                    ARM_COMPUTE_ERROR("NOT SUPPORTED!");
+            }
+        }
+        else
+        {
+            library->fill_tensor_uniform(tensor, i);
+        }
+    }
+
+    TensorType compute_target(const TensorShape &shape, DataType dt_in, DataType dt_out, ConvertPolicy policy)
+    {
+        // Create tensors
+        TensorType src = create_tensor<TensorType>(shape, dt_in, 1);
+        TensorType dst = create_tensor<TensorType>(shape, dt_out, 1);
+
+        // Create and configure function
+        FunctionType cast;
+        cast.configure(&src, &dst, policy);
+
+        ARM_COMPUTE_EXPECT(src.info()->is_resizable(), framework::LogLevel::ERRORS);
+        ARM_COMPUTE_EXPECT(dst.info()->is_resizable(), framework::LogLevel::ERRORS);
+
+        // Allocate tensors
+        src.allocator()->allocate();
+        dst.allocator()->allocate();
+
+        ARM_COMPUTE_EXPECT(!src.info()->is_resizable(), framework::LogLevel::ERRORS);
+        ARM_COMPUTE_EXPECT(!dst.info()->is_resizable(), framework::LogLevel::ERRORS);
+
+        // Fill tensors
+        fill(AccessorType(src), 0, dt_in, dt_out);
+
+        // Compute function
+        cast.run();
+
+        return dst;
+    }
+
+    SimpleTensor<T2> compute_reference(const TensorShape &shape, DataType dt_in, DataType dt_out, ConvertPolicy policy)
+    {
+        // Create reference
+        SimpleTensor<T1> src{ shape, dt_in, 1 };
+
+        // Fill reference
+        fill(src, 0, dt_in, dt_out);
+
+        return reference::depth_convert<T1, T2>(src, dt_out, policy, 0);
+    }
+
+    TensorType       _target{};
+    SimpleTensor<T2> _reference{};
+};
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* ARM_COMPUTE_TEST_CAST_FIXTURE */
diff --git a/tests/validation/reference/DepthConvertLayer.cpp b/tests/validation/reference/DepthConvertLayer.cpp
index fd2e0ae378..4d5b97b478 100644
--- a/tests/validation/reference/DepthConvertLayer.cpp
+++ b/tests/validation/reference/DepthConvertLayer.cpp
@@ -25,6 +25,9 @@
 
 #include "tests/validation/Helpers.h"
 
+#include "arm_compute/core/utils/misc/Rounding.h"
+#include "arm_compute/core/utils/misc/SaturateCast.h"
+
 #include "tests/Types.h"
 
 namespace arm_compute
@@ -35,13 +38,13 @@ namespace validation
 {
 namespace reference
 {
-template < typename T1, typename T2, typename std::enable_if < std::is_integral<T1>::value &&std::is_integral<T2>::value &&!std::is_same<T1, T2>::value, int >::type >
+template < typename T1, typename T2, typename std::enable_if < std::is_integral<T1>::value &&!std::is_same<T1, T2>::value, int >::type >
 SimpleTensor<T2> depth_convert(const SimpleTensor<T1> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift)
 {
     SimpleTensor<T2> result(src.shape(), dt_out);
 
     // Up-casting
-    if(src.data_type() <= dt_out)
+    if(element_size_from_data_type(src.data_type()) < element_size_from_data_type(dt_out))
     {
         for(int i = 0; i < src.num_elements(); ++i)
         {
@@ -54,48 +57,100 @@ SimpleTensor<T2> depth_convert(const SimpleTensor<T1> &src, DataType dt_out, Con
         for(int i = 0; i < src.num_elements(); ++i)
         {
             T1 val    = src[i] >> shift;
-            result[i] = (policy == ConvertPolicy::SATURATE) ? saturate_cast<T2>(val) : static_cast<T2>(val);
+            result[i] = (policy == ConvertPolicy::SATURATE) ? utils::cast::saturate_cast<T2>(val) : static_cast<T2>(val);
         }
     }
     return result;
 }
 
-template < typename T1, typename T2, typename std::enable_if < is_floating_point<T1>::value &&is_floating_point<T2>::value &&!std::is_same<T1, T2>::value, int >::type >
+template < typename T1, typename T2, typename std::enable_if < is_floating_point<T1>::value &&!std::is_same<T1, T2>::value, int >::type >
 SimpleTensor<T2> depth_convert(const SimpleTensor<T1> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift)
 {
     SimpleTensor<T2> result(src.shape(), dt_out);
+    ARM_COMPUTE_ERROR_ON(shift != 0);
+    ARM_COMPUTE_UNUSED(policy, shift);
 
-    const uint32_t scale = 1 << shift;
-
-    // Up-casting
-    if(src.data_type() <= dt_out)
+    // Always saturate on floats
+    for(int i = 0; i < src.num_elements(); ++i)
     {
-        for(int i = 0; i < src.num_elements(); ++i)
-        {
-            result[i] = src[i] * static_cast<T2>(scale);
-        }
-    }
-    // Down-casting
-    else
-    {
-        for(int i = 0; i < src.num_elements(); ++i)
-        {
-            T1 val    = src[i] / static_cast<T1>(scale);
-            result[i] = (policy == ConvertPolicy::SATURATE) ? saturate_cast<T2>(val) : static_cast<T2>(val);
-        }
+        T1 val    = utils::rounding::round_half_away_from_zero(src[i]);
+        result[i] = utils::cast::saturate_cast<T2>(val);
     }
     return result;
 }
 
+// U8
+template SimpleTensor<int8_t> depth_convert(const SimpleTensor<uint8_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
 template SimpleTensor<uint16_t> depth_convert(const SimpleTensor<uint8_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
 template SimpleTensor<int16_t> depth_convert(const SimpleTensor<uint8_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<uint32_t> depth_convert(const SimpleTensor<uint8_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
 template SimpleTensor<int32_t> depth_convert(const SimpleTensor<uint8_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<half> depth_convert(const SimpleTensor<uint8_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<float> depth_convert(const SimpleTensor<uint8_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+
+// S8
+template SimpleTensor<uint8_t> depth_convert(const SimpleTensor<int8_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<uint16_t> depth_convert(const SimpleTensor<int8_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<int16_t> depth_convert(const SimpleTensor<int8_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<uint32_t> depth_convert(const SimpleTensor<int8_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<int32_t> depth_convert(const SimpleTensor<int8_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<half> depth_convert(const SimpleTensor<int8_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<float> depth_convert(const SimpleTensor<int8_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+
+// U16
 template SimpleTensor<uint8_t> depth_convert(const SimpleTensor<uint16_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<int8_t> depth_convert(const SimpleTensor<uint16_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<int16_t> depth_convert(const SimpleTensor<uint16_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
 template SimpleTensor<uint32_t> depth_convert(const SimpleTensor<uint16_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<int32_t> depth_convert(const SimpleTensor<uint16_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<half> depth_convert(const SimpleTensor<uint16_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<float> depth_convert(const SimpleTensor<uint16_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+
+// S16
 template SimpleTensor<uint8_t> depth_convert(const SimpleTensor<int16_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<int8_t> depth_convert(const SimpleTensor<int16_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<uint16_t> depth_convert(const SimpleTensor<int16_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<uint32_t> depth_convert(const SimpleTensor<int16_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
 template SimpleTensor<int32_t> depth_convert(const SimpleTensor<int16_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
-template SimpleTensor<half> depth_convert(const SimpleTensor<float> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<half> depth_convert(const SimpleTensor<int16_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<float> depth_convert(const SimpleTensor<int16_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+
+// U32
+template SimpleTensor<uint8_t> depth_convert(const SimpleTensor<uint32_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<int8_t> depth_convert(const SimpleTensor<uint32_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<uint16_t> depth_convert(const SimpleTensor<uint32_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<int16_t> depth_convert(const SimpleTensor<uint32_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<int32_t> depth_convert(const SimpleTensor<uint32_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<half> depth_convert(const SimpleTensor<uint32_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<float> depth_convert(const SimpleTensor<uint32_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+
+// S32
+template SimpleTensor<uint8_t> depth_convert(const SimpleTensor<int32_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<int8_t> depth_convert(const SimpleTensor<int32_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<uint16_t> depth_convert(const SimpleTensor<int32_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<int16_t> depth_convert(const SimpleTensor<int32_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<uint32_t> depth_convert(const SimpleTensor<int32_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<half> depth_convert(const SimpleTensor<int32_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<float> depth_convert(const SimpleTensor<int32_t> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+
+// F16
+template SimpleTensor<uint8_t> depth_convert(const SimpleTensor<half> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<int8_t> depth_convert(const SimpleTensor<half> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<uint16_t> depth_convert(const SimpleTensor<half> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<int16_t> depth_convert(const SimpleTensor<half> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<uint32_t> depth_convert(const SimpleTensor<half> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<int32_t> depth_convert(const SimpleTensor<half> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
 template SimpleTensor<float> depth_convert(const SimpleTensor<half> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+
+// F32
+template SimpleTensor<uint8_t> depth_convert(const SimpleTensor<float> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<int8_t> depth_convert(const SimpleTensor<float> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<uint16_t> depth_convert(const SimpleTensor<float> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<int16_t> depth_convert(const SimpleTensor<float> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<uint32_t> depth_convert(const SimpleTensor<float> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<int32_t> depth_convert(const SimpleTensor<float> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+template SimpleTensor<half> depth_convert(const SimpleTensor<float> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
+
 } // namespace reference
 } // namespace validation
 } // namespace test
diff --git a/tests/validation/reference/DepthConvertLayer.h b/tests/validation/reference/DepthConvertLayer.h
index 5d97c73b3c..2113593e9d 100644
--- a/tests/validation/reference/DepthConvertLayer.h
+++ b/tests/validation/reference/DepthConvertLayer.h
@@ -35,10 +35,10 @@ namespace validation
 {
 namespace reference
 {
-template < typename T1, typename T2, typename std::enable_if < std::is_integral<T1>::value &&std::is_integral<T2>::value &&!std::is_same<T1, T2>::value, int >::type = 0 >
+template < typename T1, typename T2, typename std::enable_if < std::is_integral<T1>::value &&!std::is_same<T1, T2>::value, int >::type = 0 >
 SimpleTensor<T2> depth_convert(const SimpleTensor<T1> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
 
-template < typename T1, typename T2, typename std::enable_if < is_floating_point<T1>::value &&is_floating_point<T2>::value &&!std::is_same<T1, T2>::value, int >::type = 0 >
+template < typename T1, typename T2, typename std::enable_if < is_floating_point<T1>::value &&!std::is_same<T1, T2>::value, int >::type = 0 >
 SimpleTensor<T2> depth_convert(const SimpleTensor<T1> &src, DataType dt_out, ConvertPolicy policy, uint32_t shift);
 } // namespace reference
 } // namespace validation