aboutsummaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorGeorge Wort <george.wort@arm.com>2018-12-21 16:21:04 +0000
committerGeorge Wort <george.wort@arm.com>2019-01-08 11:06:01 +0000
commit5a97b28d483eefa5810f5cf57356086090c8c894 (patch)
treeb283fc76a02a7eeaadd6b275192c68cba05f4842
parent8529bd68d8a979874b8c53c0e26af411aae058f5 (diff)
downloadComputeLibrary-5a97b28d483eefa5810f5cf57356086090c8c894.tar.gz
COMPMID-1756: NEON: Implement RSqrt, Exp
Change-Id: I6b140b8868b04f7d3032a51831a80829e8e1560e Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/165590 Tested-by: bsgcomp <bsgcomp@arm.com> Reviewed-by: Anthony Barbier <anthony.barbier@arm.com> Reviewed-on: https://review.mlplatform.org/470 Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com>
-rw-r--r--arm_compute/core/NEON/NEKernels.h1
-rw-r--r--arm_compute/core/NEON/kernels/NEElementwiseUnaryKernel.h100
-rw-r--r--arm_compute/core/NEON/kernels/NEFuseBatchNormalizationKernel.h6
-rw-r--r--arm_compute/core/NEON/wrapper/intrinsics/exp.h3
-rw-r--r--arm_compute/core/NEON/wrapper/intrinsics/invsqrt.h2
-rw-r--r--arm_compute/runtime/CL/functions/CLElementwiseOperations.h4
-rw-r--r--arm_compute/runtime/NEON/NEFunctions.h1
-rw-r--r--arm_compute/runtime/NEON/functions/NEElementwiseUnaryLayer.h73
-rw-r--r--arm_compute/runtime/NEON/functions/NEFuseBatchNormalization.h6
-rw-r--r--src/core/NEON/kernels/NEElementwiseUnaryKernel.cpp206
-rw-r--r--src/runtime/NEON/functions/NEElementwiseUnaryLayer.cpp54
-rw-r--r--src/runtime/NEON/functions/NEFuseBatchNormalization.cpp6
-rw-r--r--src/runtime/NEON/functions/NEReduceMean.cpp6
-rw-r--r--tests/validation/NEON/ElementwiseExpLayer.cpp115
-rw-r--r--tests/validation/NEON/ElementwiseRsqrtLayer.cpp115
-rw-r--r--tests/validation/NEON/Tile.cpp6
16 files changed, 685 insertions, 19 deletions
diff --git a/arm_compute/core/NEON/NEKernels.h b/arm_compute/core/NEON/NEKernels.h
index 95a3e0555..6c804e771 100644
--- a/arm_compute/core/NEON/NEKernels.h
+++ b/arm_compute/core/NEON/NEKernels.h
@@ -58,6 +58,7 @@
#include "arm_compute/core/NEON/kernels/NEDirectConvolutionLayerKernel.h"
#include "arm_compute/core/NEON/kernels/NEDirectConvolutionLayerOutputStageKernel.h"
#include "arm_compute/core/NEON/kernels/NEElementwiseOperationKernel.h"
+#include "arm_compute/core/NEON/kernels/NEElementwiseUnaryKernel.h"
#include "arm_compute/core/NEON/kernels/NEErodeKernel.h"
#include "arm_compute/core/NEON/kernels/NEFastCornersKernel.h"
#include "arm_compute/core/NEON/kernels/NEFillArrayKernel.h"
diff --git a/arm_compute/core/NEON/kernels/NEElementwiseUnaryKernel.h b/arm_compute/core/NEON/kernels/NEElementwiseUnaryKernel.h
new file mode 100644
index 000000000..f632b1a93
--- /dev/null
+++ b/arm_compute/core/NEON/kernels/NEElementwiseUnaryKernel.h
@@ -0,0 +1,100 @@
+/*
+ * Copyright (c) 2018-2019 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_NEELEMENTWISEUNARYKERNEL_H__
+#define __ARM_COMPUTE_NEELEMENTWISEUNARYKERNEL_H__
+
+#include "arm_compute/core/NEON/INEKernel.h"
+#include "arm_compute/core/Types.h"
+
+namespace arm_compute
+{
+class ITensor;
+
+/** Interface for an element-wise unary operation kernel
+ *
+ * Element-wise operation is computed by:
+ * @f[ output(x) = OP(input(x))@f]
+ *
+ */
+class NEElementwiseUnaryKernel : public INEKernel
+{
+public:
+ const char *name() const override
+ {
+ return "NEElementwiseUnaryKernel";
+ }
+ /** Default constructor */
+ NEElementwiseUnaryKernel();
+ /** Prevent instances of this class from being copied (As this class contains pointers) */
+ NEElementwiseUnaryKernel(const NEElementwiseUnaryKernel &) = delete;
+ /** Prevent instances of this class from being copied (As this class contains pointers) */
+ NEElementwiseUnaryKernel &operator=(const NEElementwiseUnaryKernel &) = delete;
+ /** Allow instances of this class to be moved */
+ NEElementwiseUnaryKernel(NEElementwiseUnaryKernel &&) = default;
+ /** Allow instances of this class to be moved */
+ NEElementwiseUnaryKernel &operator=(NEElementwiseUnaryKernel &&) = default;
+ /** Default destructor */
+ ~NEElementwiseUnaryKernel() = default;
+
+ /** Static function to check if given info will lead to a valid configuration of @ref NEElementwiseUnaryKernel
+ *
+ * @param[in] op Arithmetic operation to be executed.
+ * @param[in] input First tensor input. Data types supported: F16/F32.
+ * @param[in] output Output tensor. Data types supported: Same as @p input.
+ */
+ void configure(ElementWiseUnary op, const ITensor *input, ITensor *output);
+
+ /** Static function to check if given info will lead to a valid configuration of @ref NEElementwiseUnaryKernel
+ *
+ * @param[in] op Arithmetic operation to be executed.
+ * @param[in] input First tensor input info. Data types supported: F16/F32.
+ * @param[in] output Output tensor info. Data types supported: Same as @p input.
+ *
+ * @return a Status
+ */
+ static Status validate(ElementWiseUnary op, const ITensorInfo *input, const ITensorInfo *output);
+
+ // Inherited methods overridden:
+ void run(const Window &window, const ThreadInfo &info) override;
+
+ /** Common signature for all the specialised arithmetic functions
+ *
+ * @param[in] input An input tensor. Data types supported: F16/F32.
+ * @param[out] output The output tensor. Data types supported: F16/F32.
+ * @param[in] window Region on which to execute the kernel.
+ */
+ using ElementwiseUnaryFunction = void(const ITensor *input, ITensor *output, const Window &window);
+
+protected:
+ // Inherited methods overridden:
+ static Status validate_arguments(const ITensorInfo &input, const ITensorInfo &output);
+
+ /** Function to use for the particular tensor types passed to configure() */
+ std::function<void(const ITensor *input, ITensor *output, const Window &window)> _function;
+
+ const ITensor *_input;
+ ITensor *_output;
+};
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_NEELEMENTWISEUNARYKERNEL_H__ */
diff --git a/arm_compute/core/NEON/kernels/NEFuseBatchNormalizationKernel.h b/arm_compute/core/NEON/kernels/NEFuseBatchNormalizationKernel.h
index 4d5855e6f..97d6a9f55 100644
--- a/arm_compute/core/NEON/kernels/NEFuseBatchNormalizationKernel.h
+++ b/arm_compute/core/NEON/kernels/NEFuseBatchNormalizationKernel.h
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2018 ARM Limited.
+ * Copyright (c) 2018-2019 ARM Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -14,9 +14,9 @@
* copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INNEUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * 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 NEAIM, DAMAGES OR OTHER
+ * 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.
diff --git a/arm_compute/core/NEON/wrapper/intrinsics/exp.h b/arm_compute/core/NEON/wrapper/intrinsics/exp.h
index 85aa9c739..c538ba802 100644
--- a/arm_compute/core/NEON/wrapper/intrinsics/exp.h
+++ b/arm_compute/core/NEON/wrapper/intrinsics/exp.h
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2018 ARM Limited.
+ * Copyright (c) 2018-2019 ARM Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -42,6 +42,7 @@ VEXPQ_IMPL(float32x4_t, f32)
VEXPQ_IMPL(float16x8_t, f16)
#endif // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
#undef VEXPQ_IMPL
+
} // namespace wrapper
} // namespace arm_compute
#endif /* __ARM_COMPUTE_WRAPPER_EXP_H__ */
diff --git a/arm_compute/core/NEON/wrapper/intrinsics/invsqrt.h b/arm_compute/core/NEON/wrapper/intrinsics/invsqrt.h
index 0bbf49b5c..1068cf56d 100644
--- a/arm_compute/core/NEON/wrapper/intrinsics/invsqrt.h
+++ b/arm_compute/core/NEON/wrapper/intrinsics/invsqrt.h
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2018 ARM Limited.
+ * Copyright (c) 2018-2019 ARM Limited.
*
* SPDX-License-Identifier: MIT
*
diff --git a/arm_compute/runtime/CL/functions/CLElementwiseOperations.h b/arm_compute/runtime/CL/functions/CLElementwiseOperations.h
index 4a0911ec4..1740bb0e4 100644
--- a/arm_compute/runtime/CL/functions/CLElementwiseOperations.h
+++ b/arm_compute/runtime/CL/functions/CLElementwiseOperations.h
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2018 ARM Limited.
+ * Copyright (c) 2018-2019 ARM Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -17,7 +17,7 @@
* 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, ARI SING FROM,
+ * 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.
*/
diff --git a/arm_compute/runtime/NEON/NEFunctions.h b/arm_compute/runtime/NEON/NEFunctions.h
index 53afb985f..4e198c5a4 100644
--- a/arm_compute/runtime/NEON/NEFunctions.h
+++ b/arm_compute/runtime/NEON/NEFunctions.h
@@ -57,6 +57,7 @@
#include "arm_compute/runtime/NEON/functions/NEDilate.h"
#include "arm_compute/runtime/NEON/functions/NEDirectConvolutionLayer.h"
#include "arm_compute/runtime/NEON/functions/NEElementwiseOperations.h"
+#include "arm_compute/runtime/NEON/functions/NEElementwiseUnaryLayer.h"
#include "arm_compute/runtime/NEON/functions/NEEqualizeHistogram.h"
#include "arm_compute/runtime/NEON/functions/NEErode.h"
#include "arm_compute/runtime/NEON/functions/NEFastCorners.h"
diff --git a/arm_compute/runtime/NEON/functions/NEElementwiseUnaryLayer.h b/arm_compute/runtime/NEON/functions/NEElementwiseUnaryLayer.h
new file mode 100644
index 000000000..76827903f
--- /dev/null
+++ b/arm_compute/runtime/NEON/functions/NEElementwiseUnaryLayer.h
@@ -0,0 +1,73 @@
+/*
+ * Copyright (c) 2018-2019 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_NEELEMENTWISEUNARYLAYER_H__
+#define __ARM_COMPUTE_NEELEMENTWISEUNARYLAYER_H__
+
+#include "arm_compute/runtime/NEON/INESimpleFunction.h"
+
+namespace arm_compute
+{
+class ITensor;
+
+/** Basic function to perform inverse square root on an input tensor. */
+class NERsqrtLayer : public INESimpleFunction
+{
+public:
+ /** Initialize the function
+ *
+ * @param[in] input Input tensor. Data types supported: F16/F32.
+ * @param[out] output Output tensor. Data types supported: same as @p input.
+ */
+ void configure(const ITensor *input, ITensor *output);
+ /** Static function to check if given info will lead to a valid configuration of @ref NERsqrtLayer
+ *
+ * @param[in] input First tensor input info. Data types supported: F16/F32.
+ * @param[in] output Output tensor info. Data types supported: Same as @p input.
+ *
+ * @return a status
+ */
+ static Status validate(const ITensorInfo *input, const ITensorInfo *output);
+};
+
+/** Basic function to perform exponential on an input tensor. */
+class NEExpLayer : public INESimpleFunction
+{
+public:
+ /** Initialize the function
+ *
+ * @param[in] input Input tensor. Data types supported: F16/F32.
+ * @param[out] output Output tensor. Data types supported: same as @p input.
+ */
+ void configure(const ITensor *input, ITensor *output);
+ /** Static function to check if given info will lead to a valid configuration of @ref NEExpLayer
+ *
+ * @param[in] input First tensor input info. Data types supported: F16/F32.
+ * @param[in] output Output tensor info. Data types supported: Same as @p input.
+ *
+ * @return a status
+ */
+ static Status validate(const ITensorInfo *input, const ITensorInfo *output);
+};
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_NEELEMENTWISEUNARYLAYER_H__ */
diff --git a/arm_compute/runtime/NEON/functions/NEFuseBatchNormalization.h b/arm_compute/runtime/NEON/functions/NEFuseBatchNormalization.h
index 82ed329d8..5e6286966 100644
--- a/arm_compute/runtime/NEON/functions/NEFuseBatchNormalization.h
+++ b/arm_compute/runtime/NEON/functions/NEFuseBatchNormalization.h
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2018 ARM Limited.
+ * Copyright (c) 2018-2019 ARM Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -14,9 +14,9 @@
* copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INNEUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * 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 NEAIM, DAMAGES OR OTHER
+ * 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.
diff --git a/src/core/NEON/kernels/NEElementwiseUnaryKernel.cpp b/src/core/NEON/kernels/NEElementwiseUnaryKernel.cpp
new file mode 100644
index 000000000..7ecc4d1c4
--- /dev/null
+++ b/src/core/NEON/kernels/NEElementwiseUnaryKernel.cpp
@@ -0,0 +1,206 @@
+/*
+ * Copyright (c) 2018-2019 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 "arm_compute/core/NEON/kernels/NEElementwiseUnaryKernel.h"
+
+#include "arm_compute/core/CPP/Validate.h"
+#include "arm_compute/core/Error.h"
+#include "arm_compute/core/Helpers.h"
+#include "arm_compute/core/IAccessWindow.h"
+#include "arm_compute/core/ITensor.h"
+#include "arm_compute/core/NEON/NEAsymm.h"
+#include "arm_compute/core/NEON/NEFixedPoint.h"
+#include "arm_compute/core/NEON/wrapper/wrapper.h"
+#include "arm_compute/core/TensorInfo.h"
+#include "arm_compute/core/Validate.h"
+
+#include <algorithm>
+#include <arm_neon.h>
+#include <cstdint>
+#include <map>
+#include <string>
+
+namespace arm_compute
+{
+class Coordinates;
+
+namespace
+{
+template <ElementWiseUnary op, typename ScalarType>
+inline ScalarType elementwise_op_scalar(const ScalarType &a)
+{
+ switch(op)
+ {
+ case ElementWiseUnary::RSQRT:
+ return 1 / sqrt(a);
+ case ElementWiseUnary::EXP:
+ return std::exp(a);
+ default:
+ ARM_COMPUTE_ERROR("NOT_SUPPORTED!");
+ }
+}
+
+template <ElementWiseUnary op, typename VectorType>
+inline VectorType elementwise_op(const VectorType &a)
+{
+ switch(op)
+ {
+ case ElementWiseUnary::RSQRT:
+ return wrapper::vinvsqrt(a);
+ case ElementWiseUnary::EXP:
+ return wrapper::vexpq(a);
+ default:
+ ARM_COMPUTE_ERROR("NOT_SUPPORTED!");
+ }
+}
+
+template <ElementWiseUnary op, typename ScalarType>
+void elementwise_op(const ITensor *in, ITensor *out, const Window &window)
+{
+ const int window_step_x = 16 / sizeof(ScalarType);
+ const auto window_start_x = static_cast<int>(window.x().start());
+ const auto window_end_x = static_cast<int>(window.x().end());
+
+ Window win = window;
+ win.set(Window::DimX, Window::Dimension(0, 1, 1));
+
+ Iterator input(in, win);
+ Iterator output(out, win);
+
+ execute_window_loop(win, [&](const Coordinates & id)
+ {
+ auto output_ptr = reinterpret_cast<ScalarType *>(output.ptr());
+ const auto input_ptr = reinterpret_cast<const ScalarType *>(input.ptr());
+
+ int x = window_start_x;
+ for(; x <= window_end_x - window_step_x; x += window_step_x)
+ {
+ wrapper::vstore(output_ptr + x, elementwise_op<op>(wrapper::vloadq(input_ptr + x)));
+ }
+ for(; x < window_end_x; ++x)
+ {
+ *(output_ptr + x) = elementwise_op_scalar<op>(*(input_ptr + x));
+ }
+ },
+ input, output);
+}
+
+template <ElementWiseUnary op>
+std::function<void(const ITensor *input, ITensor *output, const Window &window)>
+configure_func(const ITensor *input, ITensor *output)
+{
+ std::string function_to_call("op_");
+ function_to_call += string_from_data_type(input->info()->data_type()) + "_";
+ function_to_call += string_from_data_type(output->info()->data_type());
+
+ static std::map<std::string, NEElementwiseUnaryKernel::ElementwiseUnaryFunction *> map_function =
+ {
+ { "op_F32_F32", &elementwise_op<op, float> }
+ };
+#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+ map_function["op_F16_F16"] = &elementwise_op<op, float16_t>;
+#endif /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */
+
+ auto it = map_function.find(function_to_call);
+
+ if(it != map_function.end())
+ {
+ auto func = it->second;
+ return [func](const ITensor * input, ITensor * output, const Window & window)
+ {
+ func(input, output, window);
+ };
+ }
+ return nullptr;
+}
+} // namespace
+
+NEElementwiseUnaryKernel::NEElementwiseUnaryKernel()
+ : _function(nullptr), _input(nullptr), _output(nullptr)
+{
+}
+
+void NEElementwiseUnaryKernel::configure(ElementWiseUnary op, const ITensor *input, ITensor *output)
+{
+ ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(*input->info(), *output->info()));
+ ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
+
+ // Configure kernel window
+ const std::pair<TensorShape, ValidRegion> broadcast_pair = ITensorInfo::broadcast_shape_and_valid_region(*input->info());
+ const TensorShape &out_shape = broadcast_pair.first;
+ const ValidRegion &valid_region = broadcast_pair.second;
+
+ // Auto initialize output if not initialized
+ auto_init_if_empty(*output->info(), out_shape, 1, input->info()->data_type());
+
+ Window win = calculate_max_window(valid_region);
+
+ _input = input;
+ _output = output;
+
+ INEKernel::configure(win);
+
+ switch(op)
+ {
+ case ElementWiseUnary::RSQRT:
+ _function = configure_func<ElementWiseUnary::RSQRT>(input, output);
+ break;
+ case ElementWiseUnary::EXP:
+ _function = configure_func<ElementWiseUnary::EXP>(input, output);
+ break;
+ default:
+ ARM_COMPUTE_ERROR("NOT_SUPPORTED!");
+ }
+}
+
+Status NEElementwiseUnaryKernel::validate_arguments(const ITensorInfo &input, const ITensorInfo &output)
+{
+ ARM_COMPUTE_RETURN_ERROR_ON_CPU_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);
+ }
+
+ return Status{};
+}
+
+Status NEElementwiseUnaryKernel::validate(ElementWiseUnary op, const ITensorInfo *input, const ITensorInfo *output)
+{
+ ARM_COMPUTE_UNUSED(op);
+ ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
+ ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(*input, *output));
+ return Status{};
+}
+
+void NEElementwiseUnaryKernel::run(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);
+ ARM_COMPUTE_ERROR_ON(_function == nullptr);
+ _function(_input, _output, window);
+}
+} // namespace arm_compute
diff --git a/src/runtime/NEON/functions/NEElementwiseUnaryLayer.cpp b/src/runtime/NEON/functions/NEElementwiseUnaryLayer.cpp
new file mode 100644
index 000000000..10142c732
--- /dev/null
+++ b/src/runtime/NEON/functions/NEElementwiseUnaryLayer.cpp
@@ -0,0 +1,54 @@
+/*
+ * Copyright (c) 2018-2019 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 "arm_compute/runtime/NEON/functions/NEElementwiseUnaryLayer.h"
+
+#include "arm_compute/core/NEON/kernels/NEElementwiseUnaryKernel.h"
+#include "support/ToolchainSupport.h"
+
+#include <utility>
+
+namespace arm_compute
+{
+void NERsqrtLayer::configure(const ITensor *input, ITensor *output)
+{
+ auto k = arm_compute::support::cpp14::make_unique<NEElementwiseUnaryKernel>();
+ k->configure(ElementWiseUnary::RSQRT, input, output);
+ _kernel = std::move(k);
+}
+Status NERsqrtLayer::validate(const ITensorInfo *input, const ITensorInfo *output)
+{
+ return NEElementwiseUnaryKernel::validate(ElementWiseUnary::RSQRT, input, output);
+}
+
+void NEExpLayer::configure(const ITensor *input, ITensor *output)
+{
+ auto k = arm_compute::support::cpp14::make_unique<NEElementwiseUnaryKernel>();
+ k->configure(ElementWiseUnary::EXP, input, output);
+ _kernel = std::move(k);
+}
+Status NEExpLayer::validate(const ITensorInfo *input, const ITensorInfo *output)
+{
+ return NEElementwiseUnaryKernel::validate(ElementWiseUnary::EXP, input, output);
+}
+} // namespace arm_compute
diff --git a/src/runtime/NEON/functions/NEFuseBatchNormalization.cpp b/src/runtime/NEON/functions/NEFuseBatchNormalization.cpp
index 7f17a15cb..dc48731c4 100644
--- a/src/runtime/NEON/functions/NEFuseBatchNormalization.cpp
+++ b/src/runtime/NEON/functions/NEFuseBatchNormalization.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2018 ARM Limited.
+ * Copyright (c) 2018-2019 ARM Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -14,9 +14,9 @@
* copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INNEUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * 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 NEAIM, DAMAGES OR OTHER
+ * 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.
diff --git a/src/runtime/NEON/functions/NEReduceMean.cpp b/src/runtime/NEON/functions/NEReduceMean.cpp
index 0b022df72..c5229daae 100644
--- a/src/runtime/NEON/functions/NEReduceMean.cpp
+++ b/src/runtime/NEON/functions/NEReduceMean.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2018 ARM Limited.
+ * Copyright (c) 2018-2019 ARM Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -14,9 +14,9 @@
* copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INNEUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * 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 NEAIM, DAMAGES OR OTHER
+ * 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.
diff --git a/tests/validation/NEON/ElementwiseExpLayer.cpp b/tests/validation/NEON/ElementwiseExpLayer.cpp
new file mode 100644
index 000000000..a7e987198
--- /dev/null
+++ b/tests/validation/NEON/ElementwiseExpLayer.cpp
@@ -0,0 +1,115 @@
+/*
+ * Copyright (c) 2018-2019 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 "arm_compute/core/Types.h"
+#include "arm_compute/runtime/NEON/functions/NEElementwiseUnaryLayer.h"
+#include "arm_compute/runtime/Tensor.h"
+#include "arm_compute/runtime/TensorAllocator.h"
+#include "tests/NEON/Accessor.h"
+#include "tests/PaddingCalculator.h"
+#include "tests/datasets/ShapeDatasets.h"
+#include "tests/framework/Asserts.h"
+#include "tests/framework/Macros.h"
+#include "tests/framework/datasets/Datasets.h"
+#include "tests/validation/Validation.h"
+#include "tests/validation/fixtures/ElementWiseUnaryFixture.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace
+{
+RelativeTolerance<float> tolerance_fp32(0.000001f);
+#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+RelativeTolerance<float> tolerance_fp16(0.01f);
+#endif // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+} // namespace
+TEST_SUITE(NEON)
+TEST_SUITE(ExpLayer)
+
+DATA_TEST_CASE(Configuration, framework::DatasetMode::ALL, combine(concat(datasets::SmallShapes(), datasets::LargeShapes()), framework::dataset::make("DataType", DataType::F32)), shape, data_type)
+{
+ // Create tensors
+ Tensor src = create_tensor<Tensor>(shape, data_type);
+ Tensor dst = create_tensor<Tensor>(shape, data_type);
+
+ ARM_COMPUTE_EXPECT(src.info()->is_resizable(), framework::LogLevel::ERRORS);
+ ARM_COMPUTE_EXPECT(dst.info()->is_resizable(), framework::LogLevel::ERRORS);
+
+ // Create and configure function
+ NEExpLayer exp_layer;
+ exp_layer.configure(&src, &dst);
+
+ // Validate valid region
+ const ValidRegion valid_region = shape_to_valid_region(shape);
+ validate(src.info()->valid_region(), valid_region);
+ validate(dst.info()->valid_region(), valid_region);
+}
+
+template <typename T>
+using NEExpLayerFixture = ExpValidationFixture<Tensor, Accessor, NEExpLayer, T>;
+
+TEST_SUITE(Float)
+#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+TEST_SUITE(FP16)
+FIXTURE_DATA_TEST_CASE(RunSmall, NEExpLayerFixture<half>, framework::DatasetMode::PRECOMMIT, combine(datasets::SmallShapes(), framework::dataset::make("DataType",
+ DataType::F16)))
+{
+ // Validate output
+ validate(Accessor(_target), _reference, tolerance_fp16);
+}
+FIXTURE_DATA_TEST_CASE(RunLarge, NEExpLayerFixture<half>, framework::DatasetMode::NIGHTLY, combine(datasets::LargeShapes(), framework::dataset::make("DataType",
+ DataType::F16)))
+{
+ // Validate output
+ validate(Accessor(_target), _reference, tolerance_fp16);
+}
+
+TEST_SUITE_END() // FP16
+#endif // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+
+TEST_SUITE(FP32)
+FIXTURE_DATA_TEST_CASE(RunSmall, NEExpLayerFixture<float>, framework::DatasetMode::PRECOMMIT, combine(datasets::SmallShapes(), framework::dataset::make("DataType",
+ DataType::F32)))
+{
+ // Validate output
+ validate(Accessor(_target), _reference, tolerance_fp32);
+}
+FIXTURE_DATA_TEST_CASE(RunLarge, NEExpLayerFixture<float>, framework::DatasetMode::NIGHTLY, combine(datasets::LargeShapes(), framework::dataset::make("DataType",
+ DataType::F32)))
+{
+ // Validate output
+ validate(Accessor(_target), _reference, tolerance_fp32);
+}
+
+TEST_SUITE_END() // FP32
+TEST_SUITE_END() // Float
+
+TEST_SUITE_END() // ExpLayer
+TEST_SUITE_END() // NEON
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/NEON/ElementwiseRsqrtLayer.cpp b/tests/validation/NEON/ElementwiseRsqrtLayer.cpp
new file mode 100644
index 000000000..dbc43a8f7
--- /dev/null
+++ b/tests/validation/NEON/ElementwiseRsqrtLayer.cpp
@@ -0,0 +1,115 @@
+/*
+ * Copyright (c) 2018-2019 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 "arm_compute/core/Types.h"
+#include "arm_compute/runtime/NEON/functions/NEElementwiseUnaryLayer.h"
+#include "arm_compute/runtime/Tensor.h"
+#include "arm_compute/runtime/TensorAllocator.h"
+#include "tests/NEON/Accessor.h"
+#include "tests/PaddingCalculator.h"
+#include "tests/datasets/ShapeDatasets.h"
+#include "tests/framework/Asserts.h"
+#include "tests/framework/Macros.h"
+#include "tests/framework/datasets/Datasets.h"
+#include "tests/validation/Validation.h"
+#include "tests/validation/fixtures/ElementWiseUnaryFixture.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace
+{
+RelativeTolerance<float> tolerance_fp32(0.000001f);
+#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+RelativeTolerance<float> tolerance_fp16(0.01f);
+#endif // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+} // namespace
+TEST_SUITE(NEON)
+TEST_SUITE(RsqrtLayer)
+
+DATA_TEST_CASE(Configuration, framework::DatasetMode::ALL, combine(concat(datasets::SmallShapes(), datasets::LargeShapes()), framework::dataset::make("DataType", DataType::F32)), shape, data_type)
+{
+ // Create tensors
+ Tensor src = create_tensor<Tensor>(shape, data_type);
+ Tensor dst = create_tensor<Tensor>(shape, data_type);
+
+ ARM_COMPUTE_EXPECT(src.info()->is_resizable(), framework::LogLevel::ERRORS);
+ ARM_COMPUTE_EXPECT(dst.info()->is_resizable(), framework::LogLevel::ERRORS);
+
+ // Create and configure function
+ NERsqrtLayer exp_layer;
+ exp_layer.configure(&src, &dst);
+
+ // Validate valid region
+ const ValidRegion valid_region = shape_to_valid_region(shape);
+ validate(src.info()->valid_region(), valid_region);
+ validate(dst.info()->valid_region(), valid_region);
+}
+
+template <typename T>
+using NERsqrtLayerFixture = RsqrtValidationFixture<Tensor, Accessor, NERsqrtLayer, T>;
+
+TEST_SUITE(Float)
+#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+TEST_SUITE(FP16)
+FIXTURE_DATA_TEST_CASE(RunSmall, NERsqrtLayerFixture<half>, framework::DatasetMode::PRECOMMIT, combine(datasets::SmallShapes(), framework::dataset::make("DataType",
+ DataType::F16)))
+{
+ // Validate output
+ validate(Accessor(_target), _reference, tolerance_fp16);
+}
+FIXTURE_DATA_TEST_CASE(RunLarge, NERsqrtLayerFixture<half>, framework::DatasetMode::NIGHTLY, combine(datasets::LargeShapes(), framework::dataset::make("DataType",
+ DataType::F16)))
+{
+ // Validate output
+ validate(Accessor(_target), _reference, tolerance_fp16);
+}
+
+TEST_SUITE_END() // FP16
+#endif // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+
+TEST_SUITE(FP32)
+FIXTURE_DATA_TEST_CASE(RunSmall, NERsqrtLayerFixture<float>, framework::DatasetMode::PRECOMMIT, combine(datasets::SmallShapes(), framework::dataset::make("DataType",
+ DataType::F32)))
+{
+ // Validate output
+ validate(Accessor(_target), _reference, tolerance_fp32);
+}
+FIXTURE_DATA_TEST_CASE(RunLarge, NERsqrtLayerFixture<float>, framework::DatasetMode::NIGHTLY, combine(datasets::LargeShapes(), framework::dataset::make("DataType",
+ DataType::F32)))
+{
+ // Validate output
+ validate(Accessor(_target), _reference, tolerance_fp32);
+}
+
+TEST_SUITE_END() // FP32
+TEST_SUITE_END() // Float
+
+TEST_SUITE_END() // RsqrtLayer
+TEST_SUITE_END() // NEON
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/NEON/Tile.cpp b/tests/validation/NEON/Tile.cpp
index 209652009..f71946ad6 100644
--- a/tests/validation/NEON/Tile.cpp
+++ b/tests/validation/NEON/Tile.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2018 ARM Limited.
+ * Copyright (c) 2018-2019 ARM Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -14,9 +14,9 @@
* copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INNEUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * 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 NEAIM, DAMAGES OR OTHER
+ * 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.