Port ClTranspose to new API

Partially Resolves: COMPMID-4277 (1/2)

Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>
Change-Id: I704c2303135cbe1ba46d2fd5642c84c562204bc7
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5194
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>

6 files changed, 151 insertions, 162 deletions

diff --git a/src/core/CL/CLKernels.h b/src/core/CL/CLKernels.h
index eaac415bc4..d81b60b663 100644
--- a/src/core/CL/CLKernels.h
+++ b/src/core/CL/CLKernels.h
@@ -89,7 +89,6 @@
 #include "src/core/CL/kernels/CLStackLayerKernel.h"
 #include "src/core/CL/kernels/CLStridedSliceKernel.h"
 #include "src/core/CL/kernels/CLTileKernel.h"
-#include "src/core/CL/kernels/CLTransposeKernel.h"
 #include "src/core/CL/kernels/CLWeightsReshapeKernel.h"
 #include "src/core/CL/kernels/CLWinogradFilterTransformKernel.h"
 #include "src/core/CL/kernels/CLWinogradInputTransformKernel.h"
diff --git a/src/core/CL/kernels/CLTransposeKernel.cpp b/src/core/CL/kernels/CLTransposeKernel.cpp
deleted file mode 100644
index 56ff48be1f..0000000000
--- a/src/core/CL/kernels/CLTransposeKernel.cpp
+++ /dev/null
@@ -1,116 +0,0 @@
-/*
- * Copyright (c) 2017-2021 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.
- */
-#include "src/core/CL/kernels/CLTransposeKernel.h"
-
-#include "arm_compute/core/CL/CLHelpers.h"
-#include "arm_compute/core/CL/CLKernelLibrary.h"
-#include "arm_compute/core/CL/ICLTensor.h"
-#include "arm_compute/core/CL/OpenCL.h"
-#include "arm_compute/core/Helpers.h"
-#include "arm_compute/core/TensorInfo.h"
-#include "arm_compute/core/Utils.h"
-#include "src/core/CL/CLValidate.h"
-#include "src/core/helpers/AutoConfiguration.h"
-#include "src/core/helpers/WindowHelpers.h"
-#include "support/StringSupport.h"
-
-namespace arm_compute
-{
-namespace
-{
-TensorShape transposed_tensor_shape(const TensorShape &in)
-{
-    TensorShape  output_shape{ in };
-    const size_t w_out = in[1];
-    const size_t h_out = in[0];
-    output_shape.set(0, w_out);
-    output_shape.set(1, h_out);
-
-    return output_shape;
-}
-
-Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output)
-{
-    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
-    ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(input);
-    ARM_COMPUTE_RETURN_ERROR_ON(input->data_type() == DataType::UNKNOWN);
-
-    if(output->total_size() != 0)
-    {
-        const TensorInfo tensor_info = input->clone()->set_tensor_shape(transposed_tensor_shape(input->tensor_shape()));
-
-        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(output, &tensor_info);
-        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
-        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_QUANTIZATION_INFO(input, output);
-    }
-
-    return Status{};
-}
-} // namespace
-
-Status CLTransposeKernel::validate(const ITensorInfo *input, const ITensorInfo *output)
-{
-    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
-    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output));
-    return Status{};
-}
-
-void CLTransposeKernel::configure(const ICLTensor *input, ICLTensor *output)
-{
-    configure(CLKernelLibrary::get().get_compile_context(), input, output);
-}
-
-void CLTransposeKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output)
-{
-    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
-
-    // Output tensor auto inizialitation if not yet initialized
-    auto_init_if_empty(*output->info(), input->info()->clone()->set_tensor_shape(transposed_tensor_shape(input->info()->tensor_shape())));
-
-    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info()));
-    auto padding_info = get_padding_info({ input, output });
-
-    _input  = input;
-    _output = output;
-
-    const unsigned int vec_size_x           = adjust_vec_size(max_cl_vector_width / input->info()->element_size(), input->info()->dimension(0));
-    const int          vec_size_x_leftovers = input->info()->dimension(0) % vec_size_x;
-    const unsigned int vec_size_y           = adjust_vec_size(max_cl_vector_width / input->info()->element_size(), input->info()->dimension(1));
-    const int          vec_size_y_leftovers = input->info()->dimension(1) % vec_size_y;
-
-    CLBuildOptions build_opts;
-    build_opts.add_option("-DDATA_TYPE_IN_BYTES=" + support::cpp11::to_string(input->info()->element_size()));
-    build_opts.add_option("-DVEC_SIZE_X=" + support::cpp11::to_string(vec_size_x));
-    build_opts.add_option("-DVEC_SIZE_LEFTOVER_X=" + support::cpp11::to_string(vec_size_x_leftovers));
-    build_opts.add_option("-DVEC_SIZE_Y=" + support::cpp11::to_string(vec_size_y));
-    build_opts.add_option("-DVEC_SIZE_LEFTOVER_Y=" + support::cpp11::to_string(vec_size_y_leftovers));
-
-    _kernel = create_kernel(compile_context, "transpose", build_opts.options());
-
-    // Configure kernel window
-    Window win = calculate_max_window(*input->info(), Steps(vec_size_x, vec_size_y));
-    ICLKernel::configure_internal(win, cl::NDRange(2, 8));
-    ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
-}
-} // namespace arm_compute
\ No newline at end of file
diff --git a/src/core/gpu/cl/kernels/ClPermuteKernel.cpp b/src/core/gpu/cl/kernels/ClPermuteKernel.cpp
index 04e649b911..ffc13060a8 100644
--- a/src/core/gpu/cl/kernels/ClPermuteKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClPermuteKernel.cpp
@@ -57,7 +57,7 @@ Status validate_arguments(const ITensorInfo *src, const ITensorInfo *dst, const
     ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src, dst);
     ARM_COMPUTE_RETURN_ERROR_ON(src->data_type() == DataType::UNKNOWN);
     ARM_COMPUTE_RETURN_ERROR_ON_MSG(src->num_dimensions() < 1 || src->num_dimensions() > 4,
-                                    "Permutation upto 4-D src tensor is supported");
+                                    "Permutation up to 4-D src tensor is supported");
     ARM_COMPUTE_RETURN_ERROR_ON_MSG(perm.num_dimensions() < 1 || perm.num_dimensions() > 4,
                                     "Permutation vector size should be less than or equal to 4");
     for(const auto &p : perm)
@@ -77,17 +77,12 @@ Status validate_arguments(const ITensorInfo *src, const ITensorInfo *dst, const
 }
 } // namespace
 
-void ClPermuteKernel::configure(const ITensorInfo *src, ITensorInfo *dst, const PermutationVector &perm)
-{
-    configure(CLKernelLibrary::get().get_compile_context(), src, dst, perm);
-}
-
 void ClPermuteKernel::configure(const CLCompileContext &compile_context, const ITensorInfo *src, ITensorInfo *dst, const PermutationVector &perm)
 {
     ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);
     auto              padding_info = get_padding_info({ src, dst });
     const TensorShape dst_shape    = get_dst_shape(src, perm);
-    // Output auto inizialitation if not yet initialized
+    // Output auto initialization if not yet initialized
     auto_init_if_empty(*dst, src->clone()->set_tensor_shape(dst_shape));
 
     ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(src, dst, perm));
diff --git a/src/core/gpu/cl/kernels/ClPermuteKernel.h b/src/core/gpu/cl/kernels/ClPermuteKernel.h
index ae3996fca1..b844214595 100644
--- a/src/core/gpu/cl/kernels/ClPermuteKernel.h
+++ b/src/core/gpu/cl/kernels/ClPermuteKernel.h
@@ -38,7 +38,7 @@ namespace kernels
  *
  * Permutes given a permutation vector
  */
-class ClPermuteKernel : public ICLKernel
+class ClPermuteKernel : public IClKernel
 {
 public:
     /** Default constructor */
@@ -48,15 +48,6 @@ public:
      *
      * @note Arbitrary permutation vectors are supported with rank not greater than 4
      *
-     * @param[in] src  The src tensor info. Data types supported: All.
-     * @param[in] dst  The dst tensor info. Data types supported: Same as @p src
-     * @param[in] perm Permutation vector
-     */
-    void configure(const ITensorInfo *src, ITensorInfo *dst, const PermutationVector &perm);
-    /** Set the src and dst of the kernel.
-     *
-     * @note Arbitrary permutation vectors are supported with rank not greater than 4
-     *
      * @param[in] compile_context The compile context to be used.
      * @param[in] src             The src tensor info. Data types supported: All.
      * @param[in] dst             The dst tensor info. Data types supported: Same as @p src
diff --git a/src/core/gpu/cl/kernels/ClTransposeKernel.cpp b/src/core/gpu/cl/kernels/ClTransposeKernel.cpp
new file mode 100644
index 0000000000..704d0152cf
--- /dev/null
+++ b/src/core/gpu/cl/kernels/ClTransposeKernel.cpp
@@ -0,0 +1,119 @@
+/*
+ * Copyright (c) 2017-2021 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.
+ */
+#include "src/core/gpu/cl/kernels/ClTransposeKernel.h"
+
+#include "arm_compute/core/CL/CLHelpers.h"
+#include "arm_compute/core/CL/CLKernelLibrary.h"
+#include "arm_compute/core/CL/ICLTensor.h"
+#include "arm_compute/core/Helpers.h"
+#include "arm_compute/core/TensorInfo.h"
+#include "arm_compute/core/Utils.h"
+#include "arm_compute/core/Validate.h"
+#include "arm_compute/core/utils/misc/ShapeCalculator.h"
+#include "src/core/CL/CLValidate.h"
+#include "src/core/helpers/AutoConfiguration.h"
+#include "src/core/helpers/WindowHelpers.h"
+#include "support/Cast.h"
+#include "support/StringSupport.h"
+
+namespace arm_compute
+{
+namespace opencl
+{
+namespace kernels
+{
+void ClTransposeKernel::configure(const CLCompileContext &compile_context, const ITensorInfo *src, ITensorInfo *dst)
+{
+    ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);
+
+    // Output auto initialization if not yet initialized
+    const TensorShape dst_shape = misc::shape_calculator::compute_transposed_shape(*src);
+    auto_init_if_empty(*dst, src->clone()->set_tensor_shape(dst_shape));
+
+    ARM_COMPUTE_ERROR_THROW_ON(ClTransposeKernel::validate(src, dst));
+    auto padding_info = get_padding_info({ src, dst });
+
+    // Create kernel
+    const unsigned int vec_size_x           = adjust_vec_size(max_cl_vector_width / src->element_size(), src->dimension(0));
+    const int          vec_size_x_leftovers = src->dimension(0) % vec_size_x;
+    const unsigned int vec_size_y           = adjust_vec_size(max_cl_vector_width / src->element_size(), src->dimension(1));
+    const int          vec_size_y_leftovers = src->dimension(1) % vec_size_y;
+
+    CLBuildOptions build_opts;
+    build_opts.add_option("-DDATA_TYPE_IN_BYTES=" + support::cpp11::to_string(src->element_size()));
+    build_opts.add_option("-DVEC_SIZE_X=" + support::cpp11::to_string(vec_size_x));
+    build_opts.add_option("-DVEC_SIZE_LEFTOVER_X=" + support::cpp11::to_string(vec_size_x_leftovers));
+    build_opts.add_option("-DVEC_SIZE_Y=" + support::cpp11::to_string(vec_size_y));
+    build_opts.add_option("-DVEC_SIZE_LEFTOVER_Y=" + support::cpp11::to_string(vec_size_y_leftovers));
+
+    _kernel = create_kernel(compile_context, "transpose", build_opts.options());
+
+    // Configure kernel window
+    Window win = calculate_max_window(*src, Steps(vec_size_x, vec_size_y));
+    ICLKernel::configure_internal(win, cl::NDRange(2, 8));
+    ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
+}
+
+Status ClTransposeKernel::validate(const ITensorInfo *src, const ITensorInfo *dst)
+{
+    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src, dst);
+    ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(src);
+    ARM_COMPUTE_RETURN_ERROR_ON(src->data_type() == DataType::UNKNOWN);
+    ARM_COMPUTE_RETURN_ERROR_ON_MSG(src->num_dimensions() > 2, "Transpose up to 2-D src tensor is supported");
+
+    // Validate configured dst
+    if(dst->total_size() != 0)
+    {
+        const TensorInfo dst_info = src->clone()->set_tensor_shape(misc::shape_calculator::compute_transposed_shape(*src));
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(dst, &dst_info);
+
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_QUANTIZATION_INFO(src, dst);
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(src, dst);
+    }
+
+    return Status{};
+}
+
+void ClTransposeKernel::run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue)
+{
+    ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
+    ARM_COMPUTE_ERROR_ON_MISMATCHING_WINDOWS(ICLKernel::window(), window);
+
+    const auto src = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC));
+    auto       dst = utils::cast::polymorphic_downcast<ICLTensor *>(tensors.get_tensor(TensorType::ACL_DST));
+
+    Window slice = window.first_slice_window_2D();
+
+    do
+    {
+        unsigned int idx = 0;
+        add_2D_tensor_argument(idx, src, slice);
+        add_2D_tensor_argument(idx, dst, slice);
+        enqueue(queue, *this, slice, lws_hint());
+    }
+    while(window.slide_window_slice_2D(slice));
+}
+} // namespace kernels
+} // namespace opencl
+} // namespace arm_compute
\ No newline at end of file
diff --git a/src/core/CL/kernels/CLTransposeKernel.h b/src/core/gpu/cl/kernels/ClTransposeKernel.h
index 0c4b7b4aff..21d4fd41f5 100644
--- a/src/core/CL/kernels/CLTransposeKernel.h
+++ b/src/core/gpu/cl/kernels/ClTransposeKernel.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2020 Arm Limited.
+ * Copyright (c) 2017-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -21,44 +21,45 @@
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
-#ifndef ARM_COMPUTE_CLTRANSPOSEKERNEL_H
-#define ARM_COMPUTE_CLTRANSPOSEKERNEL_H
+#ifndef ARM_COMPUTE_CL_TRANSPOSE_KERNEL_H
+#define ARM_COMPUTE_CL_TRANSPOSE_KERNEL_H
 
-#include "src/core/CL/ICLSimple2DKernel.h"
+#include "src/core/common/Macros.h"
+#include "src/core/gpu/cl/ClCompileContext.h"
+#include "src/core/gpu/cl/IClKernel.h"
 
 namespace arm_compute
 {
-class ICLTensor;
-
-/** OpenCL kernel which transposes the elements of a matrix.
- *
- * [width, height, batch] -> [height, width, batch]
- *
- */
-class CLTransposeKernel : public ICLSimple2DKernel
+namespace opencl
+{
+namespace kernels
+{
+/** OpenCL kernel to transpose a 2D tensor. */
+class ClTransposeKernel : public IClKernel
 {
 public:
-    /** Initialise the kernel's input and output.
-     *
-     * @param[in]  input  Input tensor. Data types supported: All.
-     * @param[out] output Output tensor. Data type supported: Same as @p input
-     */
-    void configure(const ICLTensor *input, ICLTensor *output);
-    /** Initialise the kernel's input and output.
+    ClTransposeKernel() = default;
+    ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClTransposeKernel);
+    /** Set the src and dst of the kernel.
      *
-     * @param[in]  compile_context The compile context to be used.
-     * @param[in]  input           Input tensor. Data types supported: All.
-     * @param[out] output          Output tensor. Data type supported: Same as @p input
+     * @param[in] compile_context The compile context to be used.
+     * @param[in] src             The src tensor info. Data types supported: All.
+     * @param[in] dst             The dst tensor info. Data types supported: Same as @p src
      */
-    void configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output);
-    /** Static function to check if given info will lead to a valid configuration of @ref CLTransposeKernel
+    void configure(const CLCompileContext &compile_context, const ITensorInfo *src, ITensorInfo *dst);
+    /** Static function to check if given info will lead to a valid configuration of @ref ClTransposeKernel
      *
-     * @param[in] input  Input tensor. Data types supported: All.
-     * @param[in] output Output tensor. Data type supported: Same as @p input
+     * @param[in] src The src tensor info. Data types supported: All.
+     * @param[in] dst The dst tensor info. Data types supported: same as @p src.
      *
      * @return a status
      */
-    static Status validate(const ITensorInfo *input, const ITensorInfo *output);
+    static Status validate(const ITensorInfo *src, const ITensorInfo *dst);
+
+    // Inherited methods overridden:
+    void run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue) override;
 };
+} // namespace kernels
+} // namespace opencl
 } // namespace arm_compute
-#endif /* ARM_COMPUTE_CLTRANSPOSEKERNEL_H */
+#endif /*ARM_COMPUTE_CL_TRANSPOSE_KERNEL_H */