Make CLFloor and CLActivation kernels and functions state-less
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Change-Id: Icbe4e6a7c6732a59bdda0136af44c4852452dfd1
Signed-off-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/4891
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
diff --git a/src/core/CL/CLKernels.h b/src/core/CL/CLKernels.h
index 11f1d2d..76bc527 100644
--- a/src/core/CL/CLKernels.h
+++ b/src/core/CL/CLKernels.h
@@ -27,7 +27,6 @@
/* Header regrouping all the CL kernels */
#include "src/core/CL/kernels/CLAbsoluteDifferenceKernel.h"
#include "src/core/CL/kernels/CLAccumulateKernel.h"
-#include "src/core/CL/kernels/CLActivationLayerKernel.h"
#include "src/core/CL/kernels/CLArgMinMaxLayerKernel.h"
#include "src/core/CL/kernels/CLBatchNormalizationLayerKernel.h"
#include "src/core/CL/kernels/CLBatchToSpaceLayerKernel.h"
@@ -65,7 +64,6 @@
#include "src/core/CL/kernels/CLFFTScaleKernel.h"
#include "src/core/CL/kernels/CLFastCornersKernel.h"
#include "src/core/CL/kernels/CLFillBorderKernel.h"
-#include "src/core/CL/kernels/CLFloorKernel.h"
#include "src/core/CL/kernels/CLFuseBatchNormalizationKernel.h"
#include "src/core/CL/kernels/CLGEMMLowpMatrixMultiplyNativeKernel.h"
#include "src/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedKernel.h"
diff --git a/src/core/CL/kernels/CLActivationLayerKernel.h b/src/core/CL/kernels/CLActivationLayerKernel.h
deleted file mode 100644
index 821418f..0000000
--- a/src/core/CL/kernels/CLActivationLayerKernel.h
+++ /dev/null
@@ -1,77 +0,0 @@
-/*
- * Copyright (c) 2016-2020 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_CLACTIVATIONLAYERKERNEL_H
-#define ARM_COMPUTE_CLACTIVATIONLAYERKERNEL_H
-
-#include "src/core/CL/ICLKernel.h"
-
-namespace arm_compute
-{
-class ICLTensor;
-/** Interface for the activation layer kernel. */
-class CLActivationLayerKernel : public ICLKernel
-{
-public:
- /** Default constructor */
- CLActivationLayerKernel();
- /** Prevent instances of this class from being copied (As this class contains pointers) */
- CLActivationLayerKernel(const CLActivationLayerKernel &) = delete;
- /** Prevent instances of this class from being copied (As this class contains pointers) */
- CLActivationLayerKernel &operator=(const CLActivationLayerKernel &) = delete;
- /** Allow instances of this class to be moved */
- CLActivationLayerKernel(CLActivationLayerKernel &&) = default;
- /** Allow instances of this class to be moved */
- CLActivationLayerKernel &operator=(CLActivationLayerKernel &&) = default;
- /** Default destructor */
- ~CLActivationLayerKernel() = default;
- /** Set the input and output tensor.
- *
- * @note If the output tensor is a nullptr, the activation function will be performed in-place
- *
- * @param[in] compile_context The compile context to be used.
- * @param[in, out] input Source tensor. In case of @p output tensor = nullptr, this tensor will store the result
- * of the activation function. Data types supported: QASYMM8/QASYMM8_SIGNED/QSYMM16/F16/F32.
- * @param[out] output Destination tensor. Data type supported: same as @p input
- * @param[in] act_info Activation layer information.
- */
- void configure(const CLCompileContext &compile_context, ITensorInfo *input, ITensorInfo *output, ActivationLayerInfo act_info);
- /** Static function to check if given info will lead to a valid configuration of @ref CLActivationLayerKernel
- *
- * @param[in] input Source tensor info. In case of @p output tensor info = nullptr, this tensor will store the result
- * of the activation function. Data types supported: QASYMM8/QASYMM8_SIGNED/QSYMM16/F16/F32.
- * @param[in] output Destination tensor info. Data type supported: same as @p input
- * @param[in] act_info Activation layer information.
- *
- * @return a status
- */
- static Status validate(const ITensorInfo *input, const ITensorInfo *output, const ActivationLayerInfo &act_info);
-
- // Inherited methods overridden:
- void run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue) override;
-
-private:
- bool _run_in_place;
-};
-} // namespace arm_compute
-#endif /*ARM_COMPUTE_CLACTIVATIONLAYERKERNEL_H */
diff --git a/src/core/CL/kernels/CLFloorKernel.cpp b/src/core/CL/kernels/CLFloorKernel.cpp
deleted file mode 100644
index 6af1822..0000000
--- a/src/core/CL/kernels/CLFloorKernel.cpp
+++ /dev/null
@@ -1,122 +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/CLFloorKernel.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 "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
-{
-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_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::F16, DataType::F32);
-
- // Validate in case of configured output
- if(output->total_size() > 0)
- {
- ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
- ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input, output);
- }
-
- return Status{};
-}
-} // namespace
-
-CLFloorKernel::CLFloorKernel()
- : _input(nullptr), _output(nullptr)
-{
-}
-
-void CLFloorKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output)
-{
- ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
-
- // Auto initialize output
- auto_init_if_empty(*output->info(), input->info()->tensor_shape(), 1, input->info()->data_type());
-
- // Validate
- 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;
- CLBuildOptions build_opts;
- build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(input->info()->data_type()));
- build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(vec_size_x));
- build_opts.add_option("-DVEC_SIZE_LEFTOVER=" + support::cpp11::to_string(vec_size_x_leftovers));
-
- // Create kernel
- _kernel = create_kernel(compile_context, "floor_layer", build_opts.options());
-
- // Configure kernel window
- Window win = calculate_max_window(*input->info(), Steps(vec_size_x));
- ICLKernel::configure_internal(win);
- ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
-}
-
-void CLFloorKernel::configure(const ICLTensor *input, ICLTensor *output)
-{
- configure(CLKernelLibrary::get().get_compile_context(), input, output);
-}
-
-Status CLFloorKernel::validate(const ITensorInfo *input, const ITensorInfo *output)
-{
- ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output));
- return Status{};
-}
-
-void CLFloorKernel::run(const Window &window, cl::CommandQueue &queue)
-{
- ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
- ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
-
- Window collapsed = window.collapse_if_possible(ICLKernel::window(), Window::DimZ);
- Window slice = collapsed.first_slice_window_3D();
-
- do
- {
- unsigned int idx = 0;
- add_3D_tensor_argument(idx, _input, slice);
- add_3D_tensor_argument(idx, _output, slice);
- enqueue(queue, *this, slice, lws_hint());
- }
- while(collapsed.slide_window_slice_3D(slice));
-}
-} // namespace arm_compute
diff --git a/src/core/CL/kernels/CLFloorKernel.h b/src/core/CL/kernels/CLFloorKernel.h
deleted file mode 100644
index f563514..0000000
--- a/src/core/CL/kernels/CLFloorKernel.h
+++ /dev/null
@@ -1,81 +0,0 @@
-/*
- * Copyright (c) 2017-2020 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_CLFLOORKERNEL_H
-#define ARM_COMPUTE_CLFLOORKERNEL_H
-
-#include "src/core/CL/ICLKernel.h"
-
-namespace arm_compute
-{
-class ICLTensor;
-
-/** OpenCL kernel to perform a floor operation */
-class CLFloorKernel : public ICLKernel
-{
-public:
- /** Default constructor */
- CLFloorKernel();
- /** Prevent instances of this class from being copied (As this class contains pointers) */
- CLFloorKernel(const CLFloorKernel &) = delete;
- /** Prevent instances of this class from being copied (As this class contains pointers) */
- CLFloorKernel &operator=(const CLFloorKernel &) = delete;
- /** Allow instances of this class to be moved */
- CLFloorKernel(CLFloorKernel &&) = default;
- /** Allow instances of this class to be moved */
- CLFloorKernel &operator=(CLFloorKernel &&) = default;
- /** Default destructor */
- ~CLFloorKernel() = default;
- /** Set the source, destination of the kernel
- *
- * @param[in] input Source tensor. Data type supported: F16/F32.
- * @param[out] output Destination tensor. Same as @p input
- */
- void configure(const ICLTensor *input, ICLTensor *output);
-
- /** Set the source, destination of the kernel
- *
- * @param[in] compile_context The compile context to be used.
- * @param[in] input Source tensor. Data type supported: F16/F32.
- * @param[out] output Destination tensor. Same as @p input
- */
- 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 CLFloorKernel
- *
- * @param[in] input Source tensor info. Data type supported: F16/F32.
- * @param[in] output Destination tensor info. Same as @p input
- *
- * @return a status
- */
- static Status validate(const ITensorInfo *input, const ITensorInfo *output);
-
- // Inherited methods overridden:
- void run(const Window &window, cl::CommandQueue &queue) override;
-
-private:
- const ICLTensor *_input;
- ICLTensor *_output;
-};
-} // namespace arm_compute
-#endif /*ARM_COMPUTE_CLFLOORKERNEL_H */
diff --git a/src/core/CL/kernels/CLActivationLayerKernel.cpp b/src/core/gpu/cl/kernels/ClActivationKernel.cpp
similarity index 77%
rename from src/core/CL/kernels/CLActivationLayerKernel.cpp
rename to src/core/gpu/cl/kernels/ClActivationKernel.cpp
index 00b6829..f48425f 100644
--- a/src/core/CL/kernels/CLActivationLayerKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClActivationKernel.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2016-2020 Arm Limited.
+ * Copyright (c) 2016-2021 Arm Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -21,7 +21,7 @@
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*/
-#include "src/core/CL/kernels/CLActivationLayerKernel.h"
+#include "src/core/gpu/cl/kernels/ClActivationKernel.h"
#include "arm_compute/core/CL/CLCoreRuntimeContext.h"
#include "arm_compute/core/CL/CLHelpers.h"
@@ -39,12 +39,16 @@
namespace arm_compute
{
+namespace opencl
+{
+namespace kernels
+{
namespace
{
-Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, const ActivationLayerInfo &act_info)
+Status validate_arguments(const ITensorInfo *src, const ITensorInfo *dst, const ActivationLayerInfo &act_info)
{
- ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(input);
- ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED, DataType::QSYMM16, DataType::F16, DataType::F32);
+ ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(src);
+ ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(src, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED, DataType::QSYMM16, DataType::F16, DataType::F32);
static std::set<ActivationLayerInfo::ActivationFunction> quantized_supported_activations =
{
@@ -56,8 +60,8 @@
ActivationLayerInfo::ActivationFunction::HARD_SWISH,
ActivationLayerInfo::ActivationFunction::LEAKY_RELU,
};
- const DataType data_type = input->data_type();
- const QuantizationInfo &oq_info = (output != nullptr) ? output->quantization_info() : input->quantization_info();
+ const DataType data_type = src->data_type();
+ const QuantizationInfo &oq_info = (dst != nullptr) ? dst->quantization_info() : src->quantization_info();
const ActivationLayerInfo::ActivationFunction f_act = act_info.activation();
ARM_COMPUTE_RETURN_ERROR_ON_MSG(is_data_type_quantized(data_type) && (quantized_supported_activations.count(f_act) == 0),
@@ -72,41 +76,41 @@
ARM_COMPUTE_RETURN_ERROR_ON(data_type == DataType::QASYMM8_SIGNED && (f_act == ActivationLayerInfo::ActivationFunction::TANH) && (oq_info != QuantizationInfo(1.f / 128.f, 0)));
ARM_COMPUTE_RETURN_ERROR_ON(data_type == DataType::QASYMM8_SIGNED && (f_act == ActivationLayerInfo::ActivationFunction::LOGISTIC) && (oq_info != QuantizationInfo(1.f / 256.f, -128)));
- // Checks performed when output is configured
- if((output != nullptr) && (output->total_size() != 0))
+ // Checks performed when destination is configured
+ if((dst != nullptr) && (dst->total_size() != 0))
{
- ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input, output);
- ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
+ ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(src, dst);
+ ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(src, dst);
}
return Status{};
}
} // namespace
-CLActivationLayerKernel::CLActivationLayerKernel()
+ClActivationKernel::ClActivationKernel()
: _run_in_place(false)
{
}
-void CLActivationLayerKernel::configure(const CLCompileContext &compile_context, ITensorInfo *input, ITensorInfo *output, ActivationLayerInfo act_info)
+void ClActivationKernel::configure(const ClCompileContext &compile_context, ITensorInfo *src, ITensorInfo *dst, ActivationLayerInfo act_info)
{
- ARM_COMPUTE_ERROR_ON_NULLPTR(input);
+ ARM_COMPUTE_ERROR_ON_NULLPTR(src);
- auto padding_info = get_padding_info({ input, output });
+ auto padding_info = get_padding_info({ src, dst });
- _run_in_place = (output == nullptr) || (output == input);
+ _run_in_place = (dst == nullptr) || (dst == src);
- if(output != nullptr)
+ if(dst != nullptr)
{
- // Output auto inizialitation if not yet initialized
- auto_init_if_empty(*output, *input->clone());
+ // Destination auto inizialitation if not yet initialized
+ auto_init_if_empty(*dst, *src->clone());
}
- ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input, (output != nullptr) ? output : nullptr, act_info));
+ ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(src, (dst != nullptr) ? dst : nullptr, act_info));
- const unsigned int num_elems_processed_per_iteration = adjust_vec_size(16 / input->element_size(), input->dimension(0));
+ const unsigned int num_elems_processed_per_iteration = adjust_vec_size(16 / src->element_size(), src->dimension(0));
- const DataType dt = input->data_type();
+ const DataType dt = src->data_type();
float a_const = act_info.a();
float b_const = act_info.b();
@@ -125,14 +129,14 @@
build_opts.add_option("-DACT=" + lower_string(string_from_activation_func(f_act)));
build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(dt));
build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(num_elems_processed_per_iteration));
- build_opts.add_option("-DVEC_SIZE_LEFTOVER=" + support::cpp11::to_string(input->dimension(0) % num_elems_processed_per_iteration));
+ build_opts.add_option("-DVEC_SIZE_LEFTOVER=" + support::cpp11::to_string(src->dimension(0) % num_elems_processed_per_iteration));
std::string kernel_name = std::string("activation_layer");
// Set quantization info build options
if(is_quantized)
{
- const UniformQuantizationInfo iq_info = input->quantization_info().uniform();
+ const UniformQuantizationInfo iq_info = src->quantization_info().uniform();
if(!perform_activation_in_float)
{
@@ -180,10 +184,10 @@
// Set correct kernel name
kernel_name += perform_activation_in_float ? std::string("_quant_f32") : std::string("_quant");
- // Set scale and offset of the input and output if they have different quantization info
- if(output != nullptr)
+ // Set scale and offset of the source and destination if they have different quantization info
+ if(dst != nullptr)
{
- const UniformQuantizationInfo oq_info = output->quantization_info().uniform();
+ const UniformQuantizationInfo oq_info = dst->quantization_info().uniform();
if(iq_info != oq_info)
{
@@ -203,27 +207,27 @@
_kernel = create_kernel(compile_context, kernel_name, build_opts.options());
// Configure kernel window
- Window win = calculate_max_window(*input, Steps(num_elems_processed_per_iteration));
+ Window win = calculate_max_window(*src, Steps(num_elems_processed_per_iteration));
ICLKernel::configure_internal(win);
// Set config_id for enabling LWS tuning
_config_id = "activation_layer_";
_config_id += lower_string(string_from_data_type(dt));
_config_id += "_";
- _config_id += support::cpp11::to_string(input->dimension(0));
+ _config_id += support::cpp11::to_string(src->dimension(0));
_config_id += "_";
- _config_id += support::cpp11::to_string(input->dimension(1));
+ _config_id += support::cpp11::to_string(src->dimension(1));
ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
}
-Status CLActivationLayerKernel::validate(const ITensorInfo *input, const ITensorInfo *output, const ActivationLayerInfo &act_info)
+Status ClActivationKernel::validate(const ITensorInfo *src, const ITensorInfo *dst, const ActivationLayerInfo &act_info)
{
- ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, act_info));
+ ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(src, dst, act_info));
return Status{};
}
-void CLActivationLayerKernel::run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue)
+void ClActivationKernel::run_op(ITensorPack &tensors, const Window &window, ::cl::CommandQueue &queue)
{
ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
@@ -247,4 +251,6 @@
}
while(collapsed.slide_window_slice_3D(slice));
}
+} // namespace kernels
+} // namespace opencl
} // namespace arm_compute
diff --git a/src/core/gpu/cl/kernels/ClActivationKernel.h b/src/core/gpu/cl/kernels/ClActivationKernel.h
new file mode 100644
index 0000000..30adc55
--- /dev/null
+++ b/src/core/gpu/cl/kernels/ClActivationKernel.h
@@ -0,0 +1,74 @@
+/*
+ * Copyright (c) 2016-2021 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.
+ */
+#ifndef ARM_COMPUTE_CL_ACTIVATION_KERNEL_H
+#define ARM_COMPUTE_CL_ACTIVATION_KERNEL_H
+
+#include "src/core/common/Macros.h"
+#include "src/core/gpu/cl/ClCompileContext.h"
+#include "src/core/gpu/cl/IClKernel.h"
+
+namespace arm_compute
+{
+namespace opencl
+{
+namespace kernels
+{
+/** Interface for the activation kernel. */
+class ClActivationKernel : public IClKernel
+{
+public:
+ ClActivationKernel();
+ ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClActivationKernel);
+ /** Configure kernel for a given list of arguments
+ *
+ * @note If the output tensor is a nullptr, the activation function will be performed in-place
+ *
+ * @param[in] compile_context The compile context to be used.
+ * @param[in, out] src Source tensor. In case of @p dst tensor = nullptr, this tensor will store the result
+ * of the activation function. Data types supported: QASYMM8/QASYMM8_SIGNED/QSYMM16/F16/F32.
+ * @param[out] dst Destination tensor. Data type supported: same as @p src
+ * @param[in] act_info Activation layer information.
+ */
+ void configure(const ClCompileContext &compile_context, ITensorInfo *src, ITensorInfo *dst, ActivationLayerInfo act_info);
+ /** Static function to check if given info will lead to a valid configuration of @ref ClActivationKernel
+ *
+ * @param[in] src Source tensor info. In case of @p dst tensor info = nullptr, this tensor will store the result
+ * of the activation function. Data types supported: QASYMM8/QASYMM8_SIGNED/QSYMM16/F16/F32.
+ * @param[in] dst Destination tensor info. Data type supported: same as @p src
+ * @param[in] act_info Activation layer information.
+ *
+ * @return a status
+ */
+ static Status validate(const ITensorInfo *src, const ITensorInfo *dst, const ActivationLayerInfo &act_info);
+
+ // Inherited methods overridden:
+ void run_op(ITensorPack &tensors, const Window &window, ::cl::CommandQueue &queue) override;
+
+private:
+ bool _run_in_place;
+};
+} // namespace kernels
+} // namespace opencl
+} // namespace arm_compute
+#endif /* ARM_COMPUTE_CL_ACTIVATION_KERNEL_H */
diff --git a/src/core/gpu/cl/kernels/ClFloorKernel.cpp b/src/core/gpu/cl/kernels/ClFloorKernel.cpp
new file mode 100644
index 0000000..7296d40
--- /dev/null
+++ b/src/core/gpu/cl/kernels/ClFloorKernel.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/ClFloorKernel.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 "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
+{
+namespace
+{
+Status validate_arguments(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_DATA_TYPE_CHANNEL_NOT_IN(src, 1, DataType::F16, DataType::F32);
+
+ // Validate in case of configured output
+ if(dst->total_size() > 0)
+ {
+ ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(src, dst);
+ ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(src, dst);
+ }
+
+ return Status{};
+}
+} // namespace
+
+void ClFloorKernel::configure(const ClCompileContext &compile_context, const ITensorInfo *src, ITensorInfo *dst)
+{
+ ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);
+
+ // Auto initialize output
+ auto_init_if_empty(*dst, src->tensor_shape(), 1, src->data_type());
+
+ // Validate
+ ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(src, dst));
+ auto padding_info = get_padding_info({ src, dst });
+
+ 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;
+ CLBuildOptions build_opts;
+ build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(src->data_type()));
+ build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(vec_size_x));
+ build_opts.add_option("-DVEC_SIZE_LEFTOVER=" + support::cpp11::to_string(vec_size_x_leftovers));
+
+ // Create kernel
+ _kernel = create_kernel(compile_context, "floor_layer", build_opts.options());
+
+ // Configure kernel window
+ Window win = calculate_max_window(*src, Steps(vec_size_x));
+ IClKernel::configure_internal(win);
+ ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
+}
+
+Status ClFloorKernel::validate(const ITensorInfo *src, const ITensorInfo *dst)
+{
+ ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(src, dst));
+ return Status{};
+}
+
+void ClFloorKernel::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 collapsed = window.collapse_if_possible(IClKernel::window(), Window::DimZ);
+ Window slice = collapsed.first_slice_window_3D();
+
+ do
+ {
+ unsigned int idx = 0;
+ add_3D_tensor_argument(idx, src, slice);
+ add_3D_tensor_argument(idx, dst, slice);
+ enqueue(queue, *this, slice, lws_hint());
+ }
+ while(collapsed.slide_window_slice_3D(slice));
+}
+} // namespace kernels
+} // namespace opencl
+} // namespace arm_compute
diff --git a/src/core/gpu/cl/kernels/ClFloorKernel.h b/src/core/gpu/cl/kernels/ClFloorKernel.h
new file mode 100644
index 0000000..09ab801
--- /dev/null
+++ b/src/core/gpu/cl/kernels/ClFloorKernel.h
@@ -0,0 +1,66 @@
+/*
+ * 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.
+ */
+#ifndef ARM_COMPUTE_CL_FLOOR_KERNEL_H
+#define ARM_COMPUTE_CL_FLOOR_KERNEL_H
+
+#include "src/core/common/Macros.h"
+#include "src/core/gpu/cl/ClCompileContext.h"
+#include "src/core/gpu/cl/IClKernel.h"
+
+namespace arm_compute
+{
+namespace opencl
+{
+namespace kernels
+{
+/** OpenCL kernel to perform a floor operation */
+class ClFloorKernel : public IClKernel
+{
+public:
+ ClFloorKernel() = default;
+ ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClFloorKernel);
+ /** Configure kernel for a given list of arguments
+ *
+ * @param[in] compile_context The compile context to be used.
+ * @param[in] src Source tensor. Data type supported: F16/F32.
+ * @param[out] dst Destination tensor. Same as @p src
+ */
+ 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 ClFloorKernel
+ *
+ * @param[in] src Source tensor info. Data type supported: F16/F32.
+ * @param[in] dst Destination tensor info. Same as @p src
+ *
+ * @return a status
+ */
+ 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_CL_FLOOR_KERNEL_H */