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>

5 files changed, 66 insertions, 6 deletions

diff --git a/arm_compute/core/CL/CLCompileContext.h b/arm_compute/core/CL/CLCompileContext.h
index e8f2ff35da..60e0f95f83 100644
--- a/arm_compute/core/CL/CLCompileContext.h
+++ b/arm_compute/core/CL/CLCompileContext.h
@@ -310,6 +310,12 @@ public:
      */
     int32_t get_ddk_version() const;
 
+    /** Return the Gpu target of the associated device
+     *
+     * @return GPUTarget
+     */
+    GPUTarget get_gpu_target() const;
+
 private:
     /** Load program and its dependencies.
      *
diff --git a/arm_compute/core/ITensorInfo.h b/arm_compute/core/ITensorInfo.h
index 6839d697e3..ca2837e450 100644
--- a/arm_compute/core/ITensorInfo.h
+++ b/arm_compute/core/ITensorInfo.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-2021 Arm Limited.
+ * Copyright (c) 2016-2022 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -41,6 +41,11 @@ class ITensorInfo : public misc::ICloneable<ITensorInfo>
 {
 public:
     using TensorDimsState = std::vector<int>;
+    /** An id that uniquely identifies an ITensorInfo within some domain (e.g. a workload)
+     */
+    using Id = int32_t;
+    /** An invalid tensor id within a domain */
+    static constexpr Id invalid_tensor_id = -1;
     /** Get the value representing dynamic dimension state
      *
      * @return Value representing dynamic dimension state
@@ -280,7 +285,20 @@ public:
     * @return A DataLayout containing the layout data information.
     */
     virtual DataLayout data_layout() const = 0;
-
+    /** Get the workload tensor id of the tensor.
+    *
+    * @return Workload tensor id of the tensor
+    */
+    virtual Id id() const = 0;
+    /** Set the tensor id
+    */
+    virtual ITensorInfo &set_id(ITensorInfo::Id id) = 0;
+    /** Check if the tensor id is valid
+     */
+    bool has_valid_id() const
+    {
+        return id() != invalid_tensor_id;
+    }
     /** If infos are broadcast compatible tensor info's, return the broadcasted shape and the intersection of
      * the broadcasted valid regions of the tensors.
      *
diff --git a/arm_compute/core/SubTensorInfo.h b/arm_compute/core/SubTensorInfo.h
index 54836d0528..374ea5b8c6 100644
--- a/arm_compute/core/SubTensorInfo.h
+++ b/arm_compute/core/SubTensorInfo.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2021 Arm Limited.
+ * Copyright (c) 2017-2022 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -237,6 +237,17 @@ public:
         ARM_COMPUTE_ERROR_ON(_parent == nullptr);
         return _parent->data_layout();
     }
+    ITensorInfo::Id id() const override
+    {
+        ARM_COMPUTE_ERROR_ON(_parent == nullptr);
+        return _parent->id();
+    }
+    ITensorInfo &set_id(ITensorInfo::Id id) override
+    {
+        ARM_COMPUTE_ERROR_ON(_parent == nullptr);
+        _parent->set_id(id);
+        return *this;
+    }
 
 private:
     ITensorInfo    *_parent;
diff --git a/arm_compute/core/TensorInfo.h b/arm_compute/core/TensorInfo.h
index 40f9ed9806..7eb8c52d07 100644
--- a/arm_compute/core/TensorInfo.h
+++ b/arm_compute/core/TensorInfo.h
@@ -50,7 +50,7 @@ public:
     /** Allow instances of this class to be copy constructed */
     TensorInfo(const ITensorInfo &info);
     /** Allow instances of this class to be copy constructed */
-    TensorInfo(const TensorInfo &) = default;
+    TensorInfo(const TensorInfo &);
     /** Allow instances of this class to be copied */
     TensorInfo &operator=(const TensorInfo &) = default;
     /** Allow instances of this class to be move constructed */
@@ -297,6 +297,15 @@ public:
         _are_values_constant = are_values_constant;
         return *this;
     }
+    ITensorInfo::Id id() const override
+    {
+        return _id;
+    }
+    ITensorInfo &set_id(ITensorInfo::Id id) override
+    {
+        _id = id;
+        return *this;
+    }
     inline friend bool operator==(const TensorInfo &lhs, const TensorInfo &rhs);
 
 private:
@@ -320,6 +329,7 @@ private:
     QuantizationInfo _quantization_info;
     DataLayout       _data_layout;
     bool             _are_values_constant;
+    ITensorInfo::Id  _id;
 };
 
 /** Check whether two tensor info are equal.
@@ -334,7 +344,8 @@ inline bool operator==(const TensorInfo &lhs, const TensorInfo &rhs)
     return (lhs._total_size == rhs._total_size) && (lhs._offset_first_element_in_bytes == rhs._offset_first_element_in_bytes) && (lhs._strides_in_bytes == rhs._strides_in_bytes)
            && (lhs._num_channels == rhs._num_channels) && (lhs._tensor_shape == rhs._tensor_shape) && (lhs._dims_state == rhs._dims_state) && (lhs._data_type == rhs._data_type) && (lhs._format == rhs._format)
            && (lhs._is_resizable == rhs._is_resizable) && (lhs._valid_region == rhs._valid_region) && (lhs._padding == rhs._padding) && (lhs._quantization_info == rhs._quantization_info)
-           && (lhs._data_layout == rhs._data_layout) && (lhs._are_values_constant == rhs._are_values_constant);
+           && (lhs._data_layout == rhs._data_layout) && (lhs._are_values_constant == rhs._are_values_constant)
+           && (lhs._id == rhs._id);
 }
 } // namespace arm_compute
 #endif /*ARM_COMPUTE_TENSORINFO_H */
diff --git a/arm_compute/core/Types.h b/arm_compute/core/Types.h
index fc6d46c53b..b0a6475527 100644
--- a/arm_compute/core/Types.h
+++ b/arm_compute/core/Types.h
@@ -782,6 +782,20 @@ private:
     DimensionRoundingType _round_type;
 };
 
+/** Padding information for 2D operations like Conv2d */
+struct Padding2D
+{
+    Padding2D() = default;
+    Padding2D(size_t left, size_t right, size_t top, size_t bottom)
+        : left(left), right(right), top(top), bottom(bottom)
+    {
+    }
+    size_t left   = { 0 }; /**<  Padding across the width dimension on the left, in elements. */
+    size_t right  = { 0 }; /**<  Padding across the width dimension on the right, in elements. */
+    size_t top    = { 0 }; /**<  Padding across the height dimension on the top, in elements. */
+    size_t bottom = { 0 }; /**<  Padding across the height dimension on the bottom, in elements. */
+};
+
 /** Padding information for 3D operations like Conv3d */
 struct Padding3D
 {
@@ -1642,7 +1656,7 @@ public:
         LINEAR,          /**< Linear ( \f$ f(x)= ax + b \f$ ) */
         IDENTITY,        /**< Identity ( \f$ f(x)= x \f$ ) */
         HARD_SWISH,      /**< Hard-swish ( \f$ f(x) = (x \text{ReLU6}(x+3))/6 = x \min(\max(0,x+3),6)/6 \f$ ) */
-        SWISH,            /**< Swish ( \f$ f(x) = \frac{x}{1 + e^{-ax}} = x \text{logistic}(ax) \f$ ) */
+        SWISH,           /**< Swish ( \f$ f(x) = \frac{x}{1 + e^{-ax}} = x \text{logistic}(ax) \f$ ) */
         GELU             /**< GELU ( \f$ f(x) = x * 1/2 * 1 + erf(x / \sqrt{2}) \f$ ) */
     };