COMPMID-2268: Implement NEG for NEON.
Change-Id: I90c023dbea8ea12e9af677294ba576b2bfcc02a4
Signed-off-by: Usama Arif <usama.arif@arm.com>
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/184216
Tested-by: bsgcomp <bsgcomp@arm.com>
Comments-Addressed: bsgcomp <bsgcomp@arm.com>
Reviewed-by: Pablo Tello <pablo.tello@arm.com>
Reviewed-on: https://review.mlplatform.org/c/1099
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Pablo Marquez <pablo.tello@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
diff --git a/arm_compute/core/NEON/kernels/NEElementwiseUnaryKernel.h b/arm_compute/core/NEON/kernels/NEElementwiseUnaryKernel.h
index f632b1a..2c019b5 100644
--- a/arm_compute/core/NEON/kernels/NEElementwiseUnaryKernel.h
+++ b/arm_compute/core/NEON/kernels/NEElementwiseUnaryKernel.h
@@ -88,7 +88,7 @@
protected:
// Inherited methods overridden:
- static Status validate_arguments(const ITensorInfo &input, const ITensorInfo &output);
+ static Status validate_arguments(ElementWiseUnary op, 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;
diff --git a/arm_compute/core/Types.h b/arm_compute/core/Types.h
index a2dfbb7..544ebff 100644
--- a/arm_compute/core/Types.h
+++ b/arm_compute/core/Types.h
@@ -578,6 +578,7 @@
{
RSQRT, /**< Reverse square root */
EXP, /**< Exponential */
+ NEG, /**< Negate */
};
/** The normalization type used for the normalization layer */
diff --git a/arm_compute/runtime/NEON/functions/NEElementwiseUnaryLayer.h b/arm_compute/runtime/NEON/functions/NEElementwiseUnaryLayer.h
index 7682790..f4b7e89 100644
--- a/arm_compute/runtime/NEON/functions/NEElementwiseUnaryLayer.h
+++ b/arm_compute/runtime/NEON/functions/NEElementwiseUnaryLayer.h
@@ -69,5 +69,25 @@
*/
static Status validate(const ITensorInfo *input, const ITensorInfo *output);
};
+
+/** Basic function to negate an input tensor. */
+class NENegLayer : public INESimpleFunction
+{
+public:
+ /** Initialize the function
+ *
+ * @param[in] input Input tensor. Data types supported: F16/F32/S32.
+ * @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 NENegLayer
+ *
+ * @param[in] input First tensor input info. Data types supported: F16/F32/S32.
+ * @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/src/core/NEON/kernels/NEElementwiseUnaryKernel.cpp b/src/core/NEON/kernels/NEElementwiseUnaryKernel.cpp
index 34696d8..d62b165 100644
--- a/src/core/NEON/kernels/NEElementwiseUnaryKernel.cpp
+++ b/src/core/NEON/kernels/NEElementwiseUnaryKernel.cpp
@@ -55,12 +55,15 @@
return 1 / sqrt(a);
case ElementWiseUnary::EXP:
return std::exp(a);
+ case ElementWiseUnary::NEG:
+ return -a;
default:
ARM_COMPUTE_ERROR("NOT_SUPPORTED!");
}
}
-template <ElementWiseUnary op, typename VectorType>
+/* Elementwise operations that are supported for float */
+template <ElementWiseUnary op, bool is_float, typename VectorType, typename std::enable_if<is_float, int>::type = 0>
inline VectorType elementwise_op(const VectorType &a)
{
switch(op)
@@ -69,12 +72,27 @@
return wrapper::vinvsqrt(a);
case ElementWiseUnary::EXP:
return wrapper::vexpq(a);
+ case ElementWiseUnary::NEG:
+ return wrapper::vneg(a);
default:
ARM_COMPUTE_ERROR("NOT_SUPPORTED!");
}
}
-template <ElementWiseUnary op, typename ScalarType>
+/* Elementwise operations that are supported for non floats */
+template <ElementWiseUnary op, bool is_float, typename VectorType, typename std::enable_if<!is_float, int>::type = 0>
+inline VectorType elementwise_op(const VectorType &a)
+{
+ switch(op)
+ {
+ case ElementWiseUnary::NEG:
+ return wrapper::vneg(a);
+ default:
+ ARM_COMPUTE_ERROR("NOT_SUPPORTED!");
+ }
+}
+
+template <ElementWiseUnary op, typename ScalarType, bool is_float>
void elementwise_op(const ITensor *in, ITensor *out, const Window &window)
{
const int window_step_x = 16 / sizeof(ScalarType);
@@ -95,7 +113,7 @@
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)));
+ wrapper::vstore(output_ptr + x, elementwise_op<op, is_float>(wrapper::vloadq(input_ptr + x)));
}
for(; x < window_end_x; ++x)
{
@@ -115,10 +133,11 @@
static std::map<std::string, NEElementwiseUnaryKernel::ElementwiseUnaryFunction *> map_function =
{
- { "op_F32_F32", &elementwise_op<op, float> }
+ { "op_F32_F32", &elementwise_op<op, float, true> },
+ { "op_S32_S32", &elementwise_op<op, int32_t, false> },
};
#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
- map_function["op_F16_F16"] = &elementwise_op<op, float16_t>;
+ map_function["op_F16_F16"] = &elementwise_op<op, float16_t, true>;
#endif /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */
auto it = map_function.find(function_to_call);
@@ -142,7 +161,7 @@
void NEElementwiseUnaryKernel::configure(ElementWiseUnary op, const ITensor *input, ITensor *output)
{
- ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(*input->info(), *output->info()));
+ ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(op, *input->info(), *output->info()));
ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
// Configure kernel window
@@ -168,16 +187,29 @@
case ElementWiseUnary::EXP:
_function = configure_func<ElementWiseUnary::EXP>(input, output);
break;
+ case ElementWiseUnary::NEG:
+ _function = configure_func<ElementWiseUnary::NEG>(input, output);
+ break;
default:
ARM_COMPUTE_ERROR("NOT_SUPPORTED!");
}
}
-Status NEElementwiseUnaryKernel::validate_arguments(const ITensorInfo &input, const ITensorInfo &output)
+Status NEElementwiseUnaryKernel::validate_arguments(ElementWiseUnary op, 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);
-
+ switch(op)
+ {
+ case ElementWiseUnary::EXP:
+ case ElementWiseUnary::RSQRT:
+ ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(&input, 1, DataType::F16, DataType::F32);
+ break;
+ case ElementWiseUnary::NEG:
+ ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(&input, 1, DataType::F16, DataType::F32, DataType::S32);
+ break;
+ default:
+ ARM_COMPUTE_ERROR("NOT_SUPPORTED!");
+ }
// Validate in case of configured output
if(output.total_size() > 0)
{
@@ -189,9 +221,8 @@
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));
+ ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(op, *input, *output));
return Status{};
}
diff --git a/src/runtime/NEON/functions/NEElementwiseUnaryLayer.cpp b/src/runtime/NEON/functions/NEElementwiseUnaryLayer.cpp
index 10142c7..48f4975 100644
--- a/src/runtime/NEON/functions/NEElementwiseUnaryLayer.cpp
+++ b/src/runtime/NEON/functions/NEElementwiseUnaryLayer.cpp
@@ -51,4 +51,16 @@
{
return NEElementwiseUnaryKernel::validate(ElementWiseUnary::EXP, input, output);
}
+
+void NENegLayer::configure(const ITensor *input, ITensor *output)
+{
+ auto k = arm_compute::support::cpp14::make_unique<NEElementwiseUnaryKernel>();
+ k->configure(ElementWiseUnary::NEG, input, output);
+ _kernel = std::move(k);
+}
+Status NENegLayer::validate(const ITensorInfo *input, const ITensorInfo *output)
+{
+ return NEElementwiseUnaryKernel::validate(ElementWiseUnary::NEG, input, output);
+}
+
} // namespace arm_compute
diff --git a/tests/validation/NEON/ElementwiseNegation.cpp b/tests/validation/NEON/ElementwiseNegation.cpp
new file mode 100644
index 0000000..7e7c838
--- /dev/null
+++ b/tests/validation/NEON/ElementwiseNegation.cpp
@@ -0,0 +1,133 @@
+/*
+ * Copyright (c) 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(NegLayer)
+
+DATA_TEST_CASE(Configuration, framework::DatasetMode::ALL, combine(datasets::SmallShapes(), 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
+ NENegLayer neg_layer;
+ neg_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 NENegLayerFixture = NegValidationFixture<Tensor, Accessor, NENegLayer, T>;
+
+TEST_SUITE(Float)
+#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+TEST_SUITE(FP16)
+FIXTURE_DATA_TEST_CASE(RunSmall, NENegLayerFixture<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, NENegLayerFixture<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, NENegLayerFixture<float>, framework::DatasetMode::ALL, combine(datasets::SmallShapes(), framework::dataset::make("DataType",
+ DataType::F32)))
+{
+ // Validate output
+ validate(Accessor(_target), _reference, tolerance_fp32);
+}
+
+FIXTURE_DATA_TEST_CASE(RunLarge, NENegLayerFixture<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(Integer)
+TEST_SUITE(S32)
+FIXTURE_DATA_TEST_CASE(RunSmall, NENegLayerFixture<int32_t>, framework::DatasetMode::ALL, combine(datasets::SmallShapes(), framework::dataset::make("DataType",
+ DataType::S32)))
+{
+ // Validate output
+ validate(Accessor(_target), _reference);
+}
+
+FIXTURE_DATA_TEST_CASE(RunLarge, NENegLayerFixture<int32_t>, framework::DatasetMode::NIGHTLY, combine(datasets::LargeShapes(), framework::dataset::make("DataType",
+ DataType::S32)))
+{
+ // Validate output
+ validate(Accessor(_target), _reference);
+}
+TEST_SUITE_END() // S32
+TEST_SUITE_END() // Integer
+
+TEST_SUITE_END() // NegLayer
+TEST_SUITE_END() // NEON
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/fixtures/ElementWiseUnaryFixture.h b/tests/validation/fixtures/ElementWiseUnaryFixture.h
index f508bc1..ba13163 100644
--- a/tests/validation/fixtures/ElementWiseUnaryFixture.h
+++ b/tests/validation/fixtures/ElementWiseUnaryFixture.h
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2018 ARM Limited.
+ * Copyright (c) 2018-2019 ARM Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -53,7 +53,7 @@
protected:
template <typename U>
- void fill(U &&tensor, int i)
+ void fill(U &&tensor, int i, DataType data_type)
{
switch(_op)
{
@@ -69,6 +69,28 @@
library->fill(tensor, distribution, i);
break;
}
+ case ElementWiseUnary::NEG:
+ {
+ switch(data_type)
+ {
+ case DataType::F32:
+ case DataType::F16:
+ {
+ std::uniform_real_distribution<> distribution(-2.0f, 2.0f);
+ library->fill(tensor, distribution, i);
+ break;
+ }
+ case DataType::S32:
+ {
+ std::uniform_int_distribution<int32_t> distribution(-100, 100);
+ library->fill(tensor, distribution, i);
+ break;
+ }
+ default:
+ ARM_COMPUTE_ERROR("DataType for Elementwise Negation Not implemented");
+ }
+ break;
+ }
default:
ARM_COMPUTE_ERROR("Not implemented");
}
@@ -95,7 +117,7 @@
ARM_COMPUTE_EXPECT(!dst.info()->is_resizable(), framework::LogLevel::ERRORS);
// Fill tensors
- fill(AccessorType(src), 0);
+ fill(AccessorType(src), 0, data_type);
// Compute function
elwiseunary_layer.run();
@@ -109,7 +131,7 @@
SimpleTensor<T> src{ shape, data_type };
// Fill reference
- fill(src, 0);
+ fill(src, 0, data_type);
return reference::elementwise_unary<T>(src, _op);
}
@@ -140,6 +162,17 @@
ElementWiseUnaryValidationFixture<TensorType, AccessorType, FunctionType, T>::setup(shape, data_type, ElementWiseUnary::EXP);
}
};
+
+template <typename TensorType, typename AccessorType, typename FunctionType, typename T>
+class NegValidationFixture : public ElementWiseUnaryValidationFixture<TensorType, AccessorType, FunctionType, T>
+{
+public:
+ template <typename...>
+ void setup(const TensorShape &shape, DataType data_type)
+ {
+ ElementWiseUnaryValidationFixture<TensorType, AccessorType, FunctionType, T>::setup(shape, data_type, ElementWiseUnary::NEG);
+ }
+};
} // namespace validation
} // namespace test
} // namespace arm_compute
diff --git a/tests/validation/reference/ElementWiseUnary.cpp b/tests/validation/reference/ElementWiseUnary.cpp
index ae7f256..79310ea 100644
--- a/tests/validation/reference/ElementWiseUnary.cpp
+++ b/tests/validation/reference/ElementWiseUnary.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2018 ARM Limited.
+ * Copyright (c) 2018-2019 ARM Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -46,6 +46,9 @@
case ElementWiseUnary::EXP:
dst[i] = std::exp(src[i]);
break;
+ case ElementWiseUnary::NEG:
+ dst[i] = -src[i];
+ break;
default:
ARM_COMPUTE_ERROR("Not implemented");
}
@@ -56,6 +59,7 @@
template SimpleTensor<float> elementwise_unary(const SimpleTensor<float> &src, ElementWiseUnary op);
template SimpleTensor<half> elementwise_unary(const SimpleTensor<half> &src, ElementWiseUnary op);
+template SimpleTensor<int32_t> elementwise_unary(const SimpleTensor<int32_t> &src, ElementWiseUnary op);
} // namespace reference
} // namespace validation
} // namespace test