Rewrite dynamic fusion

The new version introduces the following major changes:

* Change public interface to simplify and standardize the user experience
    - Use the term "Workload" uniformly
    - Simplify operator interface to be a set of static methods:
      validate_op(), create_op()

* Separate the kernel writing into its own component (template_writer).
  This is to allow the co-development of GpuKernelWriter, and to allow
  easy replacement once GpuKernelWriter is mature.

* Optimize the core fusion algorithm used by the component graph. The
  details can be found in GpuKernelComponentGraph::fuse()

* Use Gpu instead of Cl prefixes for most of the Workload interfaces
  (except for runtime and kernel components, which have to be language specific)
  This allows the potential extension to other Gpu langauges in the
  future.

* Refactor runtime memory interface so that auxiliary tensor handling
  is separate from the user tensor passing. This is because the former
  is less stable and may require extension in the future.

* Hide source code object from the user as it is not required at the
  moment

* Deprecate the old prototype entirely by disabling it in SCons build

Resolves COMPMID-5510, COMPMID-5512, COMPMID-5513

Change-Id: If69d2362856f2de4503546b7b6cf48a525cf3079
Signed-off-by: SiCong Li <sicong.li@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8406
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Reviewed-by: Jakub Sujak <jakub.sujak@arm.com>
Reviewed-by: Viet-Hoa Do <viet-hoa.do@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>

6 files changed, 492 insertions, 0 deletions

diff --git a/arm_compute/dynamic_fusion/runtime/gpu/cl/ClWorkloadRuntime.h b/arm_compute/dynamic_fusion/runtime/gpu/cl/ClWorkloadRuntime.h
new file mode 100644
index 0000000000..326880f721
--- /dev/null
+++ b/arm_compute/dynamic_fusion/runtime/gpu/cl/ClWorkloadRuntime.h
@@ -0,0 +1,80 @@
+/*
+ * Copyright (c) 2022 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_DYNAMIC_FUSION_RUNTIME_GPU_CL_CLWORKLOADRUNTIME
+#define ARM_COMPUTE_DYNAMIC_FUSION_RUNTIME_GPU_CL_CLWORKLOADRUNTIME
+
+#include "arm_compute/dynamic_fusion/sketch/MemoryDescriptor.h"
+
+#include <map>
+#include <memory>
+
+namespace arm_compute
+{
+/** Forward declaration */
+class CLTensor;
+namespace experimental
+{
+namespace dynamic_fusion
+{
+/** Forward declaration */
+class GpuWorkloadSketch;
+
+/** OpenCL runtime to run a workload
+ */
+class ClWorkloadRuntime
+{
+public:
+    ClWorkloadRuntime();
+    ~ClWorkloadRuntime();
+    /** Configure @ref ClWorkloadRuntime
+     * @note A runtime cannot be re-configured
+     *
+     * @param[in] sketch @ref GpuWorkloadSketch with which to configure
+     */
+    Status configure(const GpuWorkloadSketch &sketch);
+    /** Perform run workload
+     * @note If the runtime is not configured, this method will not perform any action
+     *
+     * @param[in,out] tensors Tensors required by the run workloads
+     *
+     * @return Status If the run is successful
+     */
+    Status run(const std::vector<CLTensor *> &tensors);
+    /** Get auxiliary tensors of the workload and their memory requirement
+     */
+    std::vector<std::pair<CLTensor *, AuxMemoryInfo>> get_auxiliary_tensors();
+
+private:
+    /** Enqueue prepare workload
+     * @note If the runtime is not configured, this method will not perform any action
+     */
+    void prepare();
+    struct Implementation;
+    std::unique_ptr<Implementation> _impl;
+};
+
+} // namespace dynamic_fusion
+} // namespace experimental
+} // namespace arm_compute
+#endif /* ARM_COMPUTE_DYNAMIC_FUSION_RUNTIME_GPU_CL_CLWORKLOADRUNTIME */
diff --git a/arm_compute/dynamic_fusion/sketch/MemoryDescriptor.h b/arm_compute/dynamic_fusion/sketch/MemoryDescriptor.h
new file mode 100644
index 0000000000..deedf62262
--- /dev/null
+++ b/arm_compute/dynamic_fusion/sketch/MemoryDescriptor.h
@@ -0,0 +1,76 @@
+/*
+ * Copyright (c) 2022 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_DYNAMIC_FUSION_SKETCH_MEMORYDESCRIPTOR
+#define ARM_COMPUTE_DYNAMIC_FUSION_SKETCH_MEMORYDESCRIPTOR
+
+#include "arm_compute/core/ITensorInfo.h"
+
+namespace arm_compute
+{
+namespace experimental
+{
+namespace dynamic_fusion
+{
+/** Type of memory used by a workload tensor */
+enum class MemoryType
+{
+    User      = 0, /**< Memory coming directly from users, e.g. for argument tensors */
+    Auxiliary = 1, /**< Additional memory required by the workload tensor, e.g. for temporary tensors */
+};
+
+/** Memory information for tensors with @ref MemoryType::Auxiliary.
+ * This informs how much additional memory is required for auxiliary tensors
+ */
+struct AuxMemoryInfo
+{
+    AuxMemoryInfo() = default;
+
+    AuxMemoryInfo(size_t size, size_t alignment = 0) noexcept
+        : size(size),
+          alignment(alignment)
+    {
+    }
+
+    friend bool operator==(const AuxMemoryInfo &info0, const AuxMemoryInfo &info1)
+    {
+        return info0.size == info1.size && info0.alignment == info1.alignment;
+    }
+    size_t size{ 0 };      /**< Total memory size in bytes */
+    size_t alignment{ 0 }; /**< Memory alignment in bytes */
+};
+
+/** Descriptor of a workload tensor memory */
+struct MemoryDescriptor
+{
+    MemoryType    memory_type{};     /**< Memory Type*/
+    AuxMemoryInfo aux_memory_info{}; /**< Auxiliary Tensor Memory Information */
+};
+
+/** A map from @ref ITensorInfo to their corresponding @ref MemoryDescriptor */
+using MemoryDescriptorMap = std::map<ITensorInfo::Id, MemoryDescriptor>;
+
+} // namespace dynamic_fusion
+} // namespace experimental
+} // namespace arm_compute
+#endif /* ARM_COMPUTE_DYNAMIC_FUSION_SKETCH_MEMORYDESCRIPTOR */
diff --git a/arm_compute/dynamic_fusion/sketch/OperatorAttributes.h b/arm_compute/dynamic_fusion/sketch/OperatorAttributes.h
new file mode 100644
index 0000000000..22c6772926
--- /dev/null
+++ b/arm_compute/dynamic_fusion/sketch/OperatorAttributes.h
@@ -0,0 +1,65 @@
+/*
+ * Copyright (c) 2022 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_DYNAMIC_FUSION_SKETCH_OPERATORATTRIBUTES
+#define ARM_COMPUTE_DYNAMIC_FUSION_SKETCH_OPERATORATTRIBUTES
+
+#include "arm_compute/core/Size2D.h"
+#include "arm_compute/core/Types.h"
+#include <cstdint>
+
+namespace arm_compute
+{
+namespace experimental
+{
+namespace dynamic_fusion
+{
+/** Attributes are backend-agnostic parameters (in addition to the input/output tensors) of an operator.
+ */
+
+/** Conv2d attributes */
+class Conv2dAttributes
+{
+public:
+    /** Set padding */
+    Conv2dAttributes &pad(const Padding2D &pad);
+    /** Get padding */
+    Padding2D pad() const;
+    /** Set stride */
+    Conv2dAttributes &stride(const Size2D &stride);
+    /** Get stride */
+    Size2D stride() const;
+    /** Set dilation */
+    Conv2dAttributes &dilation(const Size2D &dilation);
+    /** Get dilation */
+    Size2D dilation() const;
+
+private:
+    Padding2D _pad{};              /**< Padding */
+    Size2D    _stride{ 1U, 1U };   /**< Stride */
+    Size2D    _dilation{ 1U, 1U }; /**< Dilation */
+};
+} // namespace dynamic_fusion
+} // namespace experimental
+} // namespace arm_compute
+#endif /* ARM_COMPUTE_DYNAMIC_FUSION_SKETCH_OPERATORATTRIBUTES */
diff --git a/arm_compute/dynamic_fusion/sketch/gpu/GpuWorkloadContext.h b/arm_compute/dynamic_fusion/sketch/gpu/GpuWorkloadContext.h
new file mode 100644
index 0000000000..1ee3c7e3ec
--- /dev/null
+++ b/arm_compute/dynamic_fusion/sketch/gpu/GpuWorkloadContext.h
@@ -0,0 +1,87 @@
+/*
+ * Copyright (c) 2022 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_DYNAMIC_FUSION_SKETCH_GPU_GPUWORKLOADCONTEXT
+#define ARM_COMPUTE_DYNAMIC_FUSION_SKETCH_GPU_GPUWORKLOADCONTEXT
+
+#include "arm_compute/core/GPUTarget.h"
+
+#include <memory>
+
+namespace arm_compute
+{
+/** Forward declaration */
+class CLCompileContext;
+namespace experimental
+{
+namespace dynamic_fusion
+{
+/** Gpu Information such as the Gpu target (for example, G76) */
+using GpuTarget = ::arm_compute::GPUTarget;
+
+/** Gpu Language */
+enum class GpuLanguage
+{
+    OpenCL,
+    Unknown
+};
+/** Provide context necessary for the creation and configuration of a workload
+ * e.g. gpu targets and capabilities, cl::Device for querying OpenCl extensions. Both can affect how a kernel is generated
+ *
+ * This context is shared between different operators within a sketch, and has to stay valid for the entire workload creation session.
+ * This context may also be shared between different sketches.
+ *
+ * This class only contains information for workload creation, but not for runtime (e.g. cl::Queue for enqueueing the kernels)
+ */
+class GpuWorkloadContext
+{
+public:
+    /** Constructor */
+    GpuWorkloadContext(CLCompileContext *cl_compile_context);
+    /** Allow instances of this class to be copy constructed */
+    GpuWorkloadContext(const GpuWorkloadContext &config) = default;
+    /** Allow instances of this class to be copied */
+    GpuWorkloadContext &operator=(const GpuWorkloadContext &config) = default;
+    /** Allow instances of this class to be move constructed */
+    GpuWorkloadContext(GpuWorkloadContext &&config) = default;
+    /** Allow instances of this class to be moved */
+    GpuWorkloadContext &operator=(GpuWorkloadContext &&config) = default;
+    /** Get @ref GpuLanguage of the context */
+    GpuLanguage gpu_language() const;
+    /** Get @ref GpuTarget of the context */
+    GpuTarget gpu_target() const;
+    /** Get @ref CLCompileContext
+     * If the gpu language is not OpenCL, then return nullptr
+     */
+    const CLCompileContext *cl_compile_context() const;
+
+private:
+    GpuLanguage       _gpu_language{ GpuLanguage::Unknown };
+    CLCompileContext *_cl_compile_ctx{ nullptr };
+};
+
+} // namespace dynamic_fusion
+} // namespace experimental
+} // namespace arm_compute
+
+#endif /* ARM_COMPUTE_DYNAMIC_FUSION_SKETCH_GPU_GPUWORKLOADCONTEXT */
diff --git a/arm_compute/dynamic_fusion/sketch/gpu/GpuWorkloadSketch.h b/arm_compute/dynamic_fusion/sketch/gpu/GpuWorkloadSketch.h
new file mode 100644
index 0000000000..afbe2b8d0b
--- /dev/null
+++ b/arm_compute/dynamic_fusion/sketch/gpu/GpuWorkloadSketch.h
@@ -0,0 +1,100 @@
+/*
+ * Copyright (c) 2022 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_DYNAMIC_FUSION_SKETCH_GPU_GPUWORKLOADSKETCH
+#define ARM_COMPUTE_DYNAMIC_FUSION_SKETCH_GPU_GPUWORKLOADSKETCH
+
+#include "arm_compute/core/TensorInfo.h"
+#include "arm_compute/dynamic_fusion/sketch/gpu/GpuWorkloadContext.h"
+
+#include <memory>
+
+namespace arm_compute
+{
+namespace experimental
+{
+namespace dynamic_fusion
+{
+/** A descriptor of a workload of operators
+ *
+ * A "workload" is a basic unit of computation to schedule and perform. It contains one or more operators that can be "fused" together.
+ * Note that a workload may still contain multiple kernels.
+ */
+class GpuWorkloadSketch
+{
+public:
+    /** Global context used for the creation of a workload */
+    using Context = GpuWorkloadContext;
+    /** Internal opaque implementation */
+    class Implementation;
+
+public:
+    /** Constructor
+     *
+     * @param[in] context Gpu context for the creation of a workload
+     */
+    explicit GpuWorkloadSketch(GpuWorkloadContext *context);
+    /** Destructor */
+    ~GpuWorkloadSketch();
+    /** Get the implementation */
+    Implementation &implementation();
+    /** Get the implementation */
+    const Implementation &implementation() const;
+    /** Get the gpu workload context of this sketch */
+    const GpuWorkloadContext *gpu_context() const;
+    /** Create a @ref TensorInfo associated with the workload sketch.
+     *
+     * @return TensorInfo   Newly created tensor info
+     */
+    template <typename... Args>
+    TensorInfo create_tensor_info(Args &&... args)
+    {
+        auto tensor_info = TensorInfo(std::forward<Args>(args)...);
+        tensor_info.set_id(allocate_new_tensor_id());
+        return tensor_info;
+    }
+    /** Create a @ref TensorInfo associated with the workload sketch by copying from an existing tensor info
+     * @note The newly copied tensor will have a different identity within the workload than the one copied from
+     *       To copy the identity of @p tensor_info as well, use @ref TensorInfo 's copy constructors instead
+     *
+     * @param[in] tensor_info @ref ITensorInfo to copy from
+     *
+     * @return TensorInfo   Newly created tensor info
+     */
+    TensorInfo create_tensor_info(const ITensorInfo &tensor_info);
+    /** Create a default @ref TensorInfo associated with the workload sketch
+     * It is usually used by a destination tensor whose @ref ITensorInfo is to be inferred automatically
+     *
+     * @return TensorInfo   Newly created tensor info
+     */
+    TensorInfo create_tensor_info();
+
+private:
+    ITensorInfo::Id                 allocate_new_tensor_id();
+    std::unique_ptr<Implementation> _impl; /**< Internal opaque implementation*/
+};
+
+} // namespace dynamic_fusion
+} // namespace experimental
+} // namespace arm_compute
+#endif /* ARM_COMPUTE_DYNAMIC_FUSION_SKETCH_GPU_GPUWORKLOADSKETCH */
diff --git a/arm_compute/dynamic_fusion/sketch/gpu/operators/GpuConv2d.h b/arm_compute/dynamic_fusion/sketch/gpu/operators/GpuConv2d.h
new file mode 100644
index 0000000000..fe9108d356
--- /dev/null
+++ b/arm_compute/dynamic_fusion/sketch/gpu/operators/GpuConv2d.h
@@ -0,0 +1,84 @@
+/*
+ * Copyright (c) 2022 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_DYNAMIC_FUSION_SKETCH_GPU_OPERATORS_GPUCONV2D
+#define ARM_COMPUTE_DYNAMIC_FUSION_SKETCH_GPU_OPERATORS_GPUCONV2D
+
+#include "arm_compute/core/Error.h"
+#include "arm_compute/dynamic_fusion/sketch/OperatorAttributes.h"
+
+namespace arm_compute
+{
+namespace experimental
+{
+namespace dynamic_fusion
+{
+/** Forward declaration */
+class GpuWorkloadSketch;
+
+/** Operator interface. */
+class GpuConv2d final
+{
+public:
+    /** Attributes are a set of backend-agnostic parameters that define what an operator does */
+    using Attributes = Conv2dAttributes;
+    /** Create an operator and fuse it into the workload sketch.
+     *    @note If @ref validate_op() fails, the creation also fails and may throw an error.
+     *    @note If @ref validate_op() fails, @p sketch remains unchanged and valid.
+     *
+     * Valid data type configurations:
+     * |src            |wei            |bia            |dst            |
+     * |:--------------|:--------------|:--------------|:--------------|
+     * |F16            |F16            |F16            |F16            |
+     * |F32            |F32            |F32            |F32            |
+     *
+     * Valid data layouts:
+     * - NHWC
+     *
+     * @param[in,out] sketch     Workload sketch into which the operator will be fused
+     * @param[in]     src        Source tensor
+     * @param[in]     wei        Weight tensor
+     * @param[in]     bia        (Optional) Bias tensor
+     * @param[out]    dst        Destination tensor
+     * @param[in]     attributes Operator attributes
+     */
+    static void create_op(GpuWorkloadSketch &sketch,
+                          ITensorInfo       *src,
+                          ITensorInfo       *wei,
+                          ITensorInfo       *bia,
+                          ITensorInfo       *dst,
+                          const Attributes &attributes);
+    /** Validate the operator and check if it can be fused into the workload sketch.
+     * Similar to @ref GpuConv2d::create_op()
+     */
+    static Status validate_op(const GpuWorkloadSketch &sketch,
+                              const ITensorInfo       *src,
+                              const ITensorInfo       *wei,
+                              const ITensorInfo       *bia,
+                              const ITensorInfo       *dst,
+                              const Attributes        &attributes);
+};
+} // namespace dynamic_fusion
+} // namespace experimental
+} // namespace arm_compute
+#endif /* ARM_COMPUTE_DYNAMIC_FUSION_SKETCH_GPU_OPERATORS_GPUCONV2D */