COMPMID-3384: Async support to NEStridedSliceLayerKernel kernels/functions

Signed-off-by: Michalis Spyrou <michalis.spyrou@arm.com>
Change-Id: I631c566fdf38ff0388b38e71549a1291d351f8b6
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3426
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>

6 files changed, 248 insertions, 29 deletions

diff --git a/arm_compute/core/NEON/kernels/NEStridedSliceKernel.h b/arm_compute/core/NEON/kernels/NEStridedSliceKernel.h
index 6709619a62..33132c41a5 100644
--- a/arm_compute/core/NEON/kernels/NEStridedSliceKernel.h
+++ b/arm_compute/core/NEON/kernels/NEStridedSliceKernel.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018-2019 ARM Limited.
+ * Copyright (c) 2018-2020 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -58,8 +58,8 @@ public:
      *
      * @note Supported tensor rank: up to 4
      *
-     * @param[in]  input            Source tensor. Data type supported: All
-     * @param[out] output           Destination tensor. Data type supported: Same as @p input
+     * @param[in]  input            Source tensor info. Data type supported: All
+     * @param[out] output           Destination tensor info. Data type supported: Same as @p input
      * @param[in]  starts           The starts of the dimensions of the input tensor to be sliced. The length must be of rank(input).
      * @param[in]  ends             The ends of the dimensions of the input tensor to be sliced. The length must be of rank(input).
      * @param[in]  strides          The strides of the dimensions of the input tensor to be sliced. The length must be of rank(input).
@@ -68,7 +68,7 @@ public:
      * @param[in]  shrink_axis_mask If the ith bit of shrink_axis_mask is set, it implies that the ith specification shrinks the dimensionality by 1.
      *                              A slice of size 1 starting from starts[i] in the dimension must be preserved.
      */
-    void configure(const ITensor *input, ITensor *output,
+    void configure(const ITensorInfo *input, ITensorInfo *output,
                    const Coordinates &starts, const Coordinates &ends, const BiStrides &strides,
                    int32_t begin_mask, int32_t end_mask, int32_t shrink_axis_mask);
 
@@ -91,14 +91,13 @@ public:
                            int32_t begin_mask, int32_t end_mask, int32_t shrink_axis_mask);
 
     // Inherited methods overridden:
-    void run(const Window &window, const ThreadInfo &info) override;
+    void run_op(const InputTensorMap &inputs, const OutputTensorMap &outputs,
+                const Window &window, const ThreadInfo &info) override;
 
 private:
-    const ITensor *_input;         /**< Source tensor */
-    ITensor       *_output;        /**< Destination tensor */
-    Coordinates    _starts_abs;    /**< Absolute start coordinates */
-    Coordinates    _final_strides; /**< Final strides */
-    int32_t        _shrink_mask;   /**< Shrink axis mask */
+    Coordinates _starts_abs;    /**< Absolute start coordinates */
+    Coordinates _final_strides; /**< Final strides */
+    int32_t     _shrink_mask;   /**< Shrink axis mask */
 };
 } // namespace arm_compute
 #endif /*ARM_COMPUTE_NE_STRIDED_SLICE_KERNEL_H */
diff --git a/arm_compute/runtime/NEON/functions/NESlice.h b/arm_compute/runtime/NEON/functions/NESlice.h
index 834ec27a33..2f59fc8bfe 100644
--- a/arm_compute/runtime/NEON/functions/NESlice.h
+++ b/arm_compute/runtime/NEON/functions/NESlice.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018-2019 ARM Limited.
+ * Copyright (c) 2018-2020 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -24,17 +24,72 @@
 #ifndef ARM_COMPUTE_NE_SLICE_H
 #define ARM_COMPUTE_NE_SLICE_H
 
-#include "arm_compute/runtime/NEON/INESimpleFunctionNoBorder.h"
+#include "arm_compute/runtime/IFunction.h"
+#include "arm_compute/runtime/NEON/INEOperator.h"
 
 namespace arm_compute
 {
 // Forward Declarations
 class ITensor;
 
+namespace experimental
+{
+/** Basic function to perform tensor slicing */
+class NESlice : public INEOperator
+{
+public:
+    /** Configure kernel
+     *
+     * @note Supported tensor rank: up to 4
+     * @note Start indices must be non-negative. 0 <= starts[i]
+     * @note End coordinates can be negative, which represents the number of elements before the end of that dimension.
+     * @note End indices are not inclusive unless negative.
+     *
+     * @param[in]  input  Source tensor info. Data type supported: All
+     * @param[out] output Destination tensor info. Data type supported: Same as @p input
+     * @param[in]  starts The starts of the dimensions of the input tensor to be sliced. The length must be of rank(input).
+     * @param[in]  ends   The ends of the dimensions of the input tensor to be sliced. The length must be of rank(input).
+     */
+    void configure(const ITensorInfo *input, ITensorInfo *output, const Coordinates &starts, const Coordinates &ends);
+
+    /** Static function to check if given info will lead to a valid configuration of @ref NESlice
+     *
+     * @note Supported tensor rank: up to 4
+     * @note Start indices must be non-negative. 0 <= starts[i]
+     * @note End coordinates can be negative, which represents the number of elements before the end of that dimension.
+     * @note End indices are not inclusive unless negative.
+     *
+     * @param[in] input  Source tensor info. Data type supported: All
+     * @param[in] output Destination tensor info. Data type supported: Same as @p input
+     * @param[in] starts The starts of the dimensions of the input tensor to be sliced. The length must be of rank(input).
+     * @param[in] ends   The ends of the dimensions of the input tensor to be sliced. The length must be of rank(input).
+     *
+     * @return A status
+     */
+    static Status validate(const ITensorInfo *input, const ITensorInfo *output, const Coordinates &starts, const Coordinates &ends);
+
+    // Inherited methods overridden:
+    MemoryRequirements workspace() const override;
+};
+} // namespace experimental
+
 /** Basic function to perform tensor slicing */
-class NESlice : public INESimpleFunctionNoBorder
+class NESlice : public IFunction
 {
 public:
+    /** Default Constructor */
+    NESlice();
+    /** Default Destructor */
+    ~NESlice();
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    NESlice(const NESlice &) = delete;
+    /** Default move constructor */
+    NESlice(NESlice &&);
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    NESlice &operator=(const NESlice &) = delete;
+    /** Default move assignment operator */
+    NESlice &operator=(NESlice &&);
+
     /** Configure kernel
      *
      * @note Supported tensor rank: up to 4
@@ -64,6 +119,13 @@ public:
      * @return A status
      */
     static Status validate(const ITensorInfo *input, const ITensorInfo *output, const Coordinates &starts, const Coordinates &ends);
+
+    // Inherited methods overridden:
+    void run() override;
+
+private:
+    struct Impl;
+    std::unique_ptr<Impl> _impl;
 };
 } // namespace arm_compute
 #endif /* ARM_COMPUTE_NE_SLICE_H */
diff --git a/arm_compute/runtime/NEON/functions/NEStridedSlice.h b/arm_compute/runtime/NEON/functions/NEStridedSlice.h
index 6d5e6392f5..c5bf8beea8 100644
--- a/arm_compute/runtime/NEON/functions/NEStridedSlice.h
+++ b/arm_compute/runtime/NEON/functions/NEStridedSlice.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018-2019 ARM Limited.
+ * Copyright (c) 2018-2020 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -24,17 +24,78 @@
 #ifndef ARM_COMPUTE_NE_STRIDED_SLICE_H
 #define ARM_COMPUTE_NE_STRIDED_SLICE_H
 
-#include "arm_compute/runtime/NEON/INESimpleFunction.h"
+#include "arm_compute/runtime/IFunction.h"
+#include "arm_compute/runtime/NEON/INEOperator.h"
 
 namespace arm_compute
 {
 // Forward Declarations
 class ITensor;
 
+namespace experimental
+{
+/** Basic function to run @ref NEStridedSliceKernel */
+class NEStridedSlice : public INEOperator
+{
+public:
+    /** Configure kernel
+     *
+     * @note Supported tensor rank: up to 4
+     *
+     * @param[in]  input            Source tensor info. Data type supported: All
+     * @param[out] output           Destination tensor info. Data type supported: Same as @p input
+     * @param[in]  starts           The starts of the dimensions of the input tensor to be sliced. The length must be of rank(input).
+     * @param[in]  ends             The ends of the dimensions of the input tensor to be sliced. The length must be of rank(input).
+     * @param[in]  strides          The strides of the dimensions of the input tensor to be sliced. The length must be of rank(input).
+     * @param[in]  begin_mask       (Optional) If the ith bit of begin_mask is set, starts[i] is ignored and the fullest possible range in that dimension is used instead.
+     * @param[in]  end_mask         (Optional) If the ith bit of end_mask is set, ends[i] is ignored and the fullest possible range in that dimension is used instead.
+     * @param[in]  shrink_axis_mask (Optional) If the ith bit of shrink_axis_mask is set, it implies that the ith specification shrinks the dimensionality by 1.
+     *                              A slice of size 1 starting from starts[i] in the dimension must be preserved.
+     */
+    void configure(const ITensorInfo *input, ITensorInfo *output,
+                   const Coordinates &starts, const Coordinates &ends, const BiStrides &strides,
+                   int32_t begin_mask = 0, int32_t end_mask = 0, int32_t shrink_axis_mask = 0);
+
+    /** Static function to check if given info will lead to a valid configuration of @ref NEStridedSlice
+     *
+     * @note Supported tensor rank: up to 4
+     *
+     * @param[in] input            Source tensor info. Data type supported: All
+     * @param[in] output           Destination tensor info. Data type supported: Same as @p input
+     * @param[in] starts           The starts of the dimensions of the input tensor to be sliced. The length must be of rank(input).
+     * @param[in] ends             The ends of the dimensions of the input tensor to be sliced. The length must be of rank(input).
+     * @param[in] strides          The strides of the dimensions of the input tensor to be sliced. The length must be of rank(input).
+     * @param[in] begin_mask       (Optional) If the ith bit of begin_mask is set, starts[i] is ignored and the fullest possible range in that dimension is used instead.
+     * @param[in] end_mask         (Optional) If the ith bit of end_mask is set, ends[i] is ignored and the fullest possible range in that dimension is used instead.
+     * @param[in] shrink_axis_mask (Optional) If the ith bit of shrink_axis_mask is set, it implies that the ith specification shrinks the dimensionality by 1.
+     *                             A slice of size 1 starting from starts[i] in the dimension must be preserved.
+     */
+    static Status validate(const ITensorInfo *input, const ITensorInfo *output,
+                           const Coordinates &starts, const Coordinates &ends, const BiStrides &strides,
+                           int32_t begin_mask = 0, int32_t end_mask = 0, int32_t shrink_axis_mask = 0);
+
+    // Inherited methods overridden:
+    MemoryRequirements workspace() const override;
+};
+} // namespace experimental
+
 /** Basic function to run @ref NEStridedSliceKernel */
-class NEStridedSlice : public INESimpleFunction
+class NEStridedSlice : public IFunction
 {
 public:
+    /** Default Constructor */
+    NEStridedSlice();
+    /** Default Destructor */
+    ~NEStridedSlice();
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    NEStridedSlice(const NEStridedSlice &) = delete;
+    /** Default move constructor */
+    NEStridedSlice(NEStridedSlice &&);
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    NEStridedSlice &operator=(const NEStridedSlice &) = delete;
+    /** Default move assignment operator */
+    NEStridedSlice &operator=(NEStridedSlice &&);
+
     /** Configure kernel
      *
      * @note Supported tensor rank: up to 4
@@ -70,6 +131,13 @@ public:
     static Status validate(const ITensorInfo *input, const ITensorInfo *output,
                            const Coordinates &starts, const Coordinates &ends, const BiStrides &strides,
                            int32_t begin_mask = 0, int32_t end_mask = 0, int32_t shrink_axis_mask = 0);
+
+    // Inherited methods overridden:
+    void run() override;
+
+private:
+    struct Impl;
+    std::unique_ptr<Impl> _impl;
 };
 } // namespace arm_compute
 #endif /* ARM_COMPUTE_NE_STRIDED_SLICE_H */
diff --git a/src/core/NEON/kernels/NEStridedSliceKernel.cpp b/src/core/NEON/kernels/NEStridedSliceKernel.cpp
index 15f786a521..c5661626de 100644
--- a/src/core/NEON/kernels/NEStridedSliceKernel.cpp
+++ b/src/core/NEON/kernels/NEStridedSliceKernel.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018-2019 ARM Limited.
+ * Copyright (c) 2018-2020 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -71,7 +71,7 @@ Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output,
     return Status{};
 }
 
-std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITensorInfo *output,
+std::pair<Status, Window> validate_and_configure_window(const ITensorInfo *input, ITensorInfo *output,
                                                         const Coordinates &starts, const Coordinates &ends, const BiStrides &strides,
                                                         int32_t begin_mask, int32_t end_mask, int32_t shrink_axis_mask)
 {
@@ -127,22 +127,20 @@ void strided_slice_generic(const ITensor *input, ITensor *output,
 } // namespace
 
 NEStridedSliceKernel::NEStridedSliceKernel()
-    : _input(nullptr), _output(nullptr), _starts_abs(), _final_strides(), _shrink_mask()
+    : _starts_abs(), _final_strides(), _shrink_mask()
 {
 }
 
-void NEStridedSliceKernel::configure(const ITensor *input, ITensor *output,
+void NEStridedSliceKernel::configure(const ITensorInfo *input, ITensorInfo *output,
                                      const Coordinates &starts, const Coordinates &ends, const BiStrides &strides,
                                      int32_t begin_mask, int32_t end_mask, int32_t shrink_axis_mask)
 {
     ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
-    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), starts, ends, strides, begin_mask, end_mask, shrink_axis_mask));
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input, output, starts, ends, strides, begin_mask, end_mask, shrink_axis_mask));
 
-    _input       = input;
-    _output      = output;
     _shrink_mask = shrink_axis_mask;
 
-    const TensorShape &input_shape = input->info()->tensor_shape();
+    const TensorShape &input_shape = input->tensor_shape();
 
     Coordinates ends_abs;
     std::tie(_starts_abs, ends_abs, _final_strides) = arm_compute::helpers::tensor_transform::calculate_strided_slice_coords(
@@ -151,7 +149,7 @@ void NEStridedSliceKernel::configure(const ITensor *input, ITensor *output,
                                                           begin_mask, end_mask, shrink_axis_mask);
 
     // Configure kernel window
-    auto win_config = validate_and_configure_window(input->info(), output->info(), starts, ends, strides, begin_mask, end_mask, shrink_axis_mask);
+    auto win_config = validate_and_configure_window(input, output, starts, ends, strides, begin_mask, end_mask, shrink_axis_mask);
     ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
     INEKernel::configure(win_config.second);
 }
@@ -168,13 +166,13 @@ Status NEStridedSliceKernel::validate(const ITensorInfo *input, const ITensorInf
     return Status{};
 }
 
-void NEStridedSliceKernel::run(const Window &window, const ThreadInfo &info)
+void NEStridedSliceKernel::run_op(const InputTensorMap &inputs, const OutputTensorMap &outputs, const Window &window, const ThreadInfo &info)
 {
     ARM_COMPUTE_UNUSED(info);
     ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
     ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(INEKernel::window(), window);
 
     // Dispatch kernel
-    strided_slice_generic(_input, _output, _starts_abs, _final_strides, _shrink_mask, window);
+    strided_slice_generic(inputs.at(TensorType::ACL_SRC_0), outputs.at(TensorType::ACL_DST), _starts_abs, _final_strides, _shrink_mask, window);
 }
 } // namespace arm_compute
diff --git a/src/runtime/NEON/functions/NESlice.cpp b/src/runtime/NEON/functions/NESlice.cpp
index 5da8896d6f..e431fe6cd4 100644
--- a/src/runtime/NEON/functions/NESlice.cpp
+++ b/src/runtime/NEON/functions/NESlice.cpp
@@ -33,7 +33,9 @@
 
 namespace arm_compute
 {
-void NESlice::configure(const ITensor *input, ITensor *output, const Coordinates &starts, const Coordinates &ends)
+namespace experimental
+{
+void NESlice::configure(const ITensorInfo *input, ITensorInfo *output, const Coordinates &starts, const Coordinates &ends)
 {
     ARM_COMPUTE_ERROR_ON_NULLPTR(input);
 
@@ -60,4 +62,46 @@ Status NESlice::validate(const ITensorInfo *input, const ITensorInfo *output, co
 
     return NEStridedSliceKernel::validate(input, output, starts, ends, BiStrides(), 0, slice_end_mask, 0);
 }
+
+MemoryRequirements NESlice::workspace() const
+{
+    return MemoryRequirements{};
+}
+} // namespace experimental
+
+struct NESlice::Impl
+{
+    const ITensor                         *src{ nullptr };
+    ITensor                               *dst{ nullptr };
+    std::unique_ptr<experimental::NESlice> op{ nullptr };
+};
+
+NESlice::NESlice()
+    : _impl(support::cpp14::make_unique<Impl>())
+{
+}
+NESlice::NESlice(NESlice &&) = default;
+NESlice &NESlice::operator=(NESlice &&) = default;
+NESlice::~NESlice()                     = default;
+
+Status NESlice::validate(const ITensorInfo *input, const ITensorInfo *output, const Coordinates &starts, const Coordinates &ends)
+{
+    return experimental::NESlice::validate(input, output, starts, ends);
+}
+
+void NESlice::configure(const ITensor *input, ITensor *output, const Coordinates &starts, const Coordinates &ends)
+{
+    _impl->src = input;
+    _impl->dst = output;
+    _impl->op  = arm_compute::support::cpp14::make_unique<experimental::NESlice>();
+    _impl->op->configure(input->info(), output->info(), starts, ends);
+}
+
+void NESlice::run()
+{
+    const InputTensorMap  src{ { TensorType::ACL_SRC, _impl->src } };
+    const OutputTensorMap dst{ { TensorType::ACL_DST, _impl->dst } };
+    _impl->op->run(src, dst, {});
+}
+
 } // namespace arm_compute
diff --git a/src/runtime/NEON/functions/NEStridedSlice.cpp b/src/runtime/NEON/functions/NEStridedSlice.cpp
index c9be563e17..80fafbe06b 100644
--- a/src/runtime/NEON/functions/NEStridedSlice.cpp
+++ b/src/runtime/NEON/functions/NEStridedSlice.cpp
@@ -23,13 +23,16 @@
  */
 #include "arm_compute/runtime/NEON/functions/NEStridedSlice.h"
 
+#include "arm_compute/core/ITensor.h"
 #include "arm_compute/core/NEON/kernels/NEStridedSliceKernel.h"
 #include "arm_compute/core/Types.h"
 #include "support/MemorySupport.h"
 
 namespace arm_compute
 {
-void NEStridedSlice::configure(const ITensor *input, ITensor *output,
+namespace experimental
+{
+void NEStridedSlice::configure(const ITensorInfo *input, ITensorInfo *output,
                                const Coordinates &starts, const Coordinates &ends, const BiStrides &strides,
                                int32_t begin_mask, int32_t end_mask, int32_t shrink_axis_mask)
 {
@@ -44,4 +47,49 @@ Status NEStridedSlice::validate(const ITensorInfo *input, const ITensorInfo *out
 {
     return NEStridedSliceKernel::validate(input, output, starts, ends, strides, begin_mask, end_mask, shrink_axis_mask);
 }
+
+MemoryRequirements NEStridedSlice::workspace() const
+{
+    return MemoryRequirements{};
+}
+} // namespace experimental
+
+struct NEStridedSlice::Impl
+{
+    const ITensor                                *src{ nullptr };
+    ITensor                                      *dst{ nullptr };
+    std::unique_ptr<experimental::NEStridedSlice> op{ nullptr };
+};
+
+NEStridedSlice::NEStridedSlice()
+    : _impl(support::cpp14::make_unique<Impl>())
+{
+}
+NEStridedSlice::NEStridedSlice(NEStridedSlice &&) = default;
+NEStridedSlice &NEStridedSlice::operator=(NEStridedSlice &&) = default;
+NEStridedSlice::~NEStridedSlice()                            = default;
+
+void NEStridedSlice::configure(const ITensor *input, ITensor *output,
+                               const Coordinates &starts, const Coordinates &ends, const BiStrides &strides,
+                               int32_t begin_mask, int32_t end_mask, int32_t shrink_axis_mask)
+{
+    _impl->src = input;
+    _impl->dst = output;
+    _impl->op  = arm_compute::support::cpp14::make_unique<experimental::NEStridedSlice>();
+    _impl->op->configure(input->info(), output->info(), starts, ends, strides, begin_mask, end_mask, shrink_axis_mask);
+}
+
+void NEStridedSlice::run()
+{
+    const InputTensorMap  src{ { TensorType::ACL_SRC, _impl->src } };
+    const OutputTensorMap dst{ { TensorType::ACL_DST, _impl->dst } };
+    _impl->op->run(src, dst, {});
+}
+
+Status NEStridedSlice::validate(const ITensorInfo *input, const ITensorInfo *output,
+                                const Coordinates &starts, const Coordinates &ends, const BiStrides &strides,
+                                int32_t begin_mask, int32_t end_mask, int32_t shrink_axis_mask)
+{
+    return experimental::NEStridedSlice::validate(input, output, starts, ends, strides, begin_mask, end_mask, shrink_axis_mask);
+}
 } // namespace arm_compute