Use the stable CKW API in the GPU dynamic fusion backend

- Refactor all kernels to work with the CKW stable API
- Add support for sub-tile in the op_load/op_store CKW operator
- Fix mismatch in resize
- Add comments in all kernels written with CKW to help developers
understand the structure of the code
- Add texture image support in depthwise convolution written with CKW
- Add support for different block sizes in depthwise convolution
- Remove the use of the dynamic fusion helper functions.
- Add support for floor in the op_unary() of CKW

Resolves: COMPMID-6708, COMPMID-6743, COMPMID-6530

Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Signed-off-by: Gunes Bayir <gunes.bayir@arm.com>
Signed-off-by: Viet-Hoa Do <viet-hoa.do@arm.com>
Signed-off-by: Jakub Sujak <jakub.sujak@arm.com>

Change-Id: I8104ce4d04a3138a1aeb0b84940e1f1c89e76069
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/10914
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Jakub Sujak <jakub.sujak@arm.com>
Reviewed-by: Gunes Bayir <gunes.bayir@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>

1 files changed, 14 insertions, 152 deletions

diff --git a/tests/validation/dynamic_fusion/gpu/cl/Cast.cpp b/tests/validation/dynamic_fusion/gpu/cl/Cast.cpp
index cb6c8c52f6..4ef359e74d 100644
--- a/tests/validation/dynamic_fusion/gpu/cl/Cast.cpp
+++ b/tests/validation/dynamic_fusion/gpu/cl/Cast.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2022 Arm Limited.
+ * Copyright (c) 2022-2024 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -42,84 +42,27 @@ namespace validation
 namespace
 {
 // Tolerance
-constexpr AbsoluteTolerance<float> one_tolerance(1);
 constexpr AbsoluteTolerance<float> zero_tolerance(0);
 
 /** Input data sets **/
-// QASYMM8
-const auto CastQASYMM8toF32Dataset = combine(framework::dataset::make("DataType", DataType::QASYMM8), framework::dataset::make("DataType", DataType::F32));
-
-// 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));
+
+class DFConvertPolicies final : public framework::dataset::ContainerDataset<std::vector<ConvertPolicy>>
+{
+public:
+    DFConvertPolicies()
+        : ContainerDataset("ConvertPolicy",
+    {
+        ConvertPolicy::WRAP
+    })
+    {
+    }
+};
 } // namespace
 
 TEST_SUITE(CL)
@@ -127,18 +70,6 @@ TEST_SUITE(DYNAMIC_FUSION)
 TEST_SUITE(CAST)
 
 template <typename T>
-using DynamicFusionCLCastToU8Fixture = DynamicFusionCastValidationFixture<CLTensor, CLAccessor, GpuCast, T, uint8_t>;
-template <typename T>
-using DynamicFusionCLCastToS8Fixture = DynamicFusionCastValidationFixture<CLTensor, CLAccessor, GpuCast, T, int8_t>;
-template <typename T>
-using DynamicFusionCLCastToU16Fixture = DynamicFusionCastValidationFixture<CLTensor, CLAccessor, GpuCast, T, uint16_t>;
-template <typename T>
-using DynamicFusionCLCastToS16Fixture = DynamicFusionCastValidationFixture<CLTensor, CLAccessor, GpuCast, T, int16_t>;
-template <typename T>
-using DynamicFusionCLCastToU32Fixture = DynamicFusionCastValidationFixture<CLTensor, CLAccessor, GpuCast, T, uint32_t>;
-template <typename T>
-using DynamicFusionCLCastToS32Fixture = DynamicFusionCastValidationFixture<CLTensor, CLAccessor, GpuCast, T, int32_t>;
-template <typename T>
 using DynamicFusionCLCastToF16Fixture = DynamicFusionCastValidationFixture<CLTensor, CLAccessor, GpuCast, T, half>;
 template <typename T>
 using DynamicFusionCLCastToF32Fixture = DynamicFusionCastValidationFixture<CLTensor, CLAccessor, GpuCast, T, float>;
@@ -146,85 +77,16 @@ using DynamicFusionCLCastToF32Fixture = DynamicFusionCastValidationFixture<CLTen
 #define CAST_SUITE(NAME, idt, odt, type, dataset, tolerance)                                                                     \
     TEST_SUITE(NAME)                                                                                                             \
     FIXTURE_DATA_TEST_CASE(RunSmall, type, framework::DatasetMode::PRECOMMIT, combine(combine(datasets::SmallShapes(), dataset), \
-                                                                                      datasets::ConvertPolicies()))              \
+                                                                                              DFConvertPolicies()))              \
     {                                                                                                                            \
         validate(CLAccessor(_target), _reference, tolerance);                                                                    \
     }                                                                                                                            \
     TEST_SUITE_END()
 
-// QASYMM8
-CAST_SUITE(QASYMM8_to_F32, DataType::QASYMM8, DataType::F32, DynamicFusionCLCastToF32Fixture<uint8_t>, CastQASYMM8toF32Dataset, zero_tolerance)
-
-// U8
-CAST_SUITE(U8_to_S8, DataType::U8, DataType::S8, DynamicFusionCLCastToS8Fixture<uint8_t>, CastU8toS8Dataset, zero_tolerance)
-CAST_SUITE(U8_to_U16, DataType::U8, DataType::U16, DynamicFusionCLCastToU16Fixture<uint8_t>, CastU8toU16Dataset, zero_tolerance)
-CAST_SUITE(U8_to_S16, DataType::U8, DataType::S16, DynamicFusionCLCastToS16Fixture<uint8_t>, CastU8toS16Dataset, zero_tolerance)
-CAST_SUITE(U8_to_U32, DataType::U8, DataType::U32, DynamicFusionCLCastToU32Fixture<uint8_t>, CastU8toU32Dataset, zero_tolerance)
-CAST_SUITE(U8_to_S32, DataType::U8, DataType::S32, DynamicFusionCLCastToS32Fixture<uint8_t>, CastU8toS32Dataset, zero_tolerance)
-CAST_SUITE(U8_to_F16, DataType::U8, DataType::F16, DynamicFusionCLCastToF16Fixture<uint8_t>, CastU8toF16Dataset, zero_tolerance)
-CAST_SUITE(U8_to_F32, DataType::U8, DataType::F32, DynamicFusionCLCastToF32Fixture<uint8_t>, CastU8toF32Dataset, zero_tolerance)
-
-// S8
-CAST_SUITE(S8_to_U8, DataType::S8, DataType::U8, DynamicFusionCLCastToU8Fixture<int8_t>, CastS8toU8Dataset, zero_tolerance)
-CAST_SUITE(S8_to_U16, DataType::S8, DataType::U16, DynamicFusionCLCastToU16Fixture<int8_t>, CastS8toU16Dataset, zero_tolerance)
-CAST_SUITE(S8_to_S16, DataType::S8, DataType::S16, DynamicFusionCLCastToS16Fixture<int8_t>, CastS8toS16Dataset, zero_tolerance)
-CAST_SUITE(S8_to_U32, DataType::S8, DataType::U32, DynamicFusionCLCastToU32Fixture<int8_t>, CastS8toU32Dataset, zero_tolerance)
-CAST_SUITE(S8_to_S32, DataType::S8, DataType::S32, DynamicFusionCLCastToS32Fixture<int8_t>, CastS8toS32Dataset, zero_tolerance)
-CAST_SUITE(S8_to_F16, DataType::S8, DataType::F16, DynamicFusionCLCastToF16Fixture<int8_t>, CastS8toF16Dataset, zero_tolerance)
-CAST_SUITE(S8_to_F32, DataType::S8, DataType::F32, DynamicFusionCLCastToF32Fixture<int8_t>, CastS8toF32Dataset, zero_tolerance)
-
-// U16
-CAST_SUITE(U16_to_U8, DataType::U16, DataType::U8, DynamicFusionCLCastToU8Fixture<uint16_t>, CastU16toU8Dataset, zero_tolerance)
-CAST_SUITE(U16_to_S8, DataType::U16, DataType::S8, DynamicFusionCLCastToS8Fixture<uint16_t>, CastU16toS8Dataset, zero_tolerance)
-CAST_SUITE(U16_to_S16, DataType::U16, DataType::S16, DynamicFusionCLCastToS16Fixture<uint16_t>, CastU16toS16Dataset, zero_tolerance)
-CAST_SUITE(U16_to_U32, DataType::U16, DataType::U32, DynamicFusionCLCastToU32Fixture<uint16_t>, CastU16toU32Dataset, zero_tolerance)
-CAST_SUITE(U16_to_S32, DataType::U16, DataType::S32, DynamicFusionCLCastToS32Fixture<uint16_t>, CastU16toS32Dataset, zero_tolerance)
-CAST_SUITE(U16_to_F16, DataType::U16, DataType::F16, DynamicFusionCLCastToF16Fixture<uint16_t>, CastU16toF16Dataset, zero_tolerance)
-CAST_SUITE(U16_to_F32, DataType::U16, DataType::F32, DynamicFusionCLCastToF32Fixture<uint16_t>, CastU16toF32Dataset, zero_tolerance)
-
-// S16
-CAST_SUITE(S16_to_U8, DataType::S16, DataType::U8, DynamicFusionCLCastToU8Fixture<int16_t>, CastS16toU8Dataset, zero_tolerance)
-CAST_SUITE(S16_to_S8, DataType::S16, DataType::S8, DynamicFusionCLCastToS8Fixture<int16_t>, CastS16toS8Dataset, zero_tolerance)
-CAST_SUITE(S16_to_U16, DataType::S16, DataType::U16, DynamicFusionCLCastToU16Fixture<int16_t>, CastS16toU16Dataset, zero_tolerance)
-CAST_SUITE(S16_to_U32, DataType::S16, DataType::U32, DynamicFusionCLCastToU32Fixture<int16_t>, CastS16toU32Dataset, zero_tolerance)
-CAST_SUITE(S16_to_S32, DataType::S16, DataType::S32, DynamicFusionCLCastToS32Fixture<int16_t>, CastS16toS32Dataset, zero_tolerance)
-CAST_SUITE(S16_to_F16, DataType::S16, DataType::F16, DynamicFusionCLCastToF16Fixture<int16_t>, CastS16toF16Dataset, zero_tolerance)
-CAST_SUITE(S16_to_F32, DataType::S16, DataType::F32, DynamicFusionCLCastToF32Fixture<int16_t>, CastS16toF32Dataset, zero_tolerance)
-
-// U32
-CAST_SUITE(U32_to_U8, DataType::U32, DataType::U8, DynamicFusionCLCastToU8Fixture<uint32_t>, CastU32toU8Dataset, zero_tolerance)
-CAST_SUITE(U32_to_S8, DataType::U32, DataType::S8, DynamicFusionCLCastToS8Fixture<uint32_t>, CastU32toS8Dataset, zero_tolerance)
-CAST_SUITE(U32_to_U16, DataType::U32, DataType::U16, DynamicFusionCLCastToU16Fixture<uint32_t>, CastU32toU16Dataset, zero_tolerance)
-CAST_SUITE(U32_to_S16, DataType::U32, DataType::S16, DynamicFusionCLCastToS16Fixture<uint32_t>, CastU32toS16Dataset, zero_tolerance)
-CAST_SUITE(U32_to_S32, DataType::U32, DataType::S32, DynamicFusionCLCastToS32Fixture<uint32_t>, CastU32toS32Dataset, zero_tolerance)
-CAST_SUITE(U32_to_F16, DataType::U32, DataType::F16, DynamicFusionCLCastToF16Fixture<uint32_t>, CastU32toF16Dataset, zero_tolerance)
-CAST_SUITE(U32_to_F32, DataType::U32, DataType::F32, DynamicFusionCLCastToF32Fixture<uint32_t>, CastU32toF32Dataset, zero_tolerance)
-
-// S32
-CAST_SUITE(S32_to_U8, DataType::S32, DataType::U8, DynamicFusionCLCastToU8Fixture<int32_t>, CastS32toU8Dataset, zero_tolerance)
-CAST_SUITE(S32_to_S8, DataType::S32, DataType::S8, DynamicFusionCLCastToS8Fixture<int32_t>, CastS32toS8Dataset, zero_tolerance)
-CAST_SUITE(S32_to_U16, DataType::S32, DataType::U16, DynamicFusionCLCastToU16Fixture<int32_t>, CastS32toU16Dataset, zero_tolerance)
-CAST_SUITE(S32_to_S16, DataType::S32, DataType::S16, DynamicFusionCLCastToS16Fixture<int32_t>, CastS32toS16Dataset, zero_tolerance)
-CAST_SUITE(S32_to_U32, DataType::S32, DataType::U32, DynamicFusionCLCastToU32Fixture<int32_t>, CastS32toU32Dataset, zero_tolerance)
-CAST_SUITE(S32_to_F16, DataType::S32, DataType::F16, DynamicFusionCLCastToF16Fixture<int32_t>, CastS32toF16Dataset, zero_tolerance)
-CAST_SUITE(S32_to_F32, DataType::S32, DataType::F32, DynamicFusionCLCastToF32Fixture<int32_t>, CastS32toF32Dataset, zero_tolerance)
-
 // F16
-CAST_SUITE(F16_to_U8, DataType::F16, DataType::U8, DynamicFusionCLCastToU8Fixture<half>, CastF16toU8Dataset, one_tolerance)
-CAST_SUITE(F16_to_S8, DataType::F16, DataType::S8, DynamicFusionCLCastToS8Fixture<half>, CastF16toS8Dataset, one_tolerance)
-CAST_SUITE(F16_to_U16, DataType::F16, DataType::U16, DynamicFusionCLCastToU16Fixture<half>, CastF16toU16Dataset, one_tolerance)
-CAST_SUITE(F16_to_S16, DataType::F16, DataType::S16, DynamicFusionCLCastToS16Fixture<half>, CastF16toS16Dataset, one_tolerance)
-CAST_SUITE(F16_to_U32, DataType::F16, DataType::U32, DynamicFusionCLCastToU32Fixture<half>, CastF16toU32Dataset, one_tolerance)
-CAST_SUITE(F16_to_S32, DataType::F16, DataType::S32, DynamicFusionCLCastToS32Fixture<half>, CastF16toS32Dataset, one_tolerance)
 CAST_SUITE(F16_to_F32, DataType::F16, DataType::F32, DynamicFusionCLCastToF32Fixture<half>, CastF16toF32Dataset, zero_tolerance)
 
 // F32
-CAST_SUITE(F32_to_U8, DataType::F32, DataType::U8, DynamicFusionCLCastToU8Fixture<float>, CastF32toU8Dataset, one_tolerance)
-CAST_SUITE(F32_to_S8, DataType::F32, DataType::S8, DynamicFusionCLCastToS8Fixture<float>, CastF32toS8Dataset, one_tolerance)
-CAST_SUITE(F32_to_U16, DataType::F32, DataType::U16, DynamicFusionCLCastToU16Fixture<float>, CastF32toU16Dataset, one_tolerance)
-CAST_SUITE(F32_to_S16, DataType::F32, DataType::S16, DynamicFusionCLCastToS16Fixture<float>, CastF32toS16Dataset, one_tolerance)
-CAST_SUITE(F32_to_U32, DataType::F32, DataType::U32, DynamicFusionCLCastToU32Fixture<float>, CastF32toU32Dataset, one_tolerance)
-CAST_SUITE(F32_to_S32, DataType::F32, DataType::S32, DynamicFusionCLCastToS32Fixture<float>, CastF32toS32Dataset, one_tolerance)
 CAST_SUITE(F32_to_F16, DataType::F32, DataType::F16, DynamicFusionCLCastToF16Fixture<float>, CastF32toF16Dataset, zero_tolerance)
 
 TEST_SUITE_END() // CAST