COMPMID-1008: Fix Doxygen issues
Change-Id: Ie73d8771f85d1f5b059f3a56f1bbd73c98e94a38
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/124723
Reviewed-by: Michalis Spyrou <michalis.spyrou@arm.com>
Tested-by: Jenkins <bsgcomp@arm.com>
diff --git a/arm_compute/core/NEON/kernels/NEAccumulateKernel.h b/arm_compute/core/NEON/kernels/NEAccumulateKernel.h
index ad5a16c..82a4199 100644
--- a/arm_compute/core/NEON/kernels/NEAccumulateKernel.h
+++ b/arm_compute/core/NEON/kernels/NEAccumulateKernel.h
@@ -101,6 +101,7 @@
void run(const Window &window, const ThreadInfo &info) override;
};
#else /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */
+/** Interface for the accumulate weighted kernel using F16 */
using NEAccumulateWeightedFP16Kernel = NEAccumulateWeightedKernel;
#endif /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */
diff --git a/arm_compute/core/NEON/kernels/NEBatchNormalizationLayerKernel.h b/arm_compute/core/NEON/kernels/NEBatchNormalizationLayerKernel.h
index 63eb739..2408a66 100644
--- a/arm_compute/core/NEON/kernels/NEBatchNormalizationLayerKernel.h
+++ b/arm_compute/core/NEON/kernels/NEBatchNormalizationLayerKernel.h
@@ -47,7 +47,7 @@
NEBatchNormalizationLayerKernel &operator=(const NEBatchNormalizationLayerKernel &) = delete;
/** Default Move Constructor. */
NEBatchNormalizationLayerKernel(NEBatchNormalizationLayerKernel &&) = default;
- /** Default move assignment operator. */
+ /** Default move assignment operator */
NEBatchNormalizationLayerKernel &operator=(NEBatchNormalizationLayerKernel &&) = default;
/** Default destructor */
~NEBatchNormalizationLayerKernel() = default;
diff --git a/arm_compute/core/NEON/kernels/NEBox3x3Kernel.h b/arm_compute/core/NEON/kernels/NEBox3x3Kernel.h
index 9c13955..2f93fd2 100644
--- a/arm_compute/core/NEON/kernels/NEBox3x3Kernel.h
+++ b/arm_compute/core/NEON/kernels/NEBox3x3Kernel.h
@@ -51,7 +51,7 @@
};
#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
-/** NEON kernel to perform a Box 3x3 filter using F16 simd
+/** NEON kernel to perform a Box 3x3 filter for FP16 datatype
*/
class NEBox3x3FP16Kernel : public NEBox3x3Kernel
{
@@ -64,6 +64,7 @@
void run(const Window &window, const ThreadInfo &info) override;
};
#else /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */
+/** NEON kernel to perform a Box 3x3 filter for FP16 datatype */
using NEBox3x3FP16Kernel = NEBox3x3Kernel;
#endif /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */
} // namespace arm_compute
diff --git a/arm_compute/core/NEON/kernels/NECannyEdgeKernel.h b/arm_compute/core/NEON/kernels/NECannyEdgeKernel.h
index 401b9e4..58ef175 100644
--- a/arm_compute/core/NEON/kernels/NECannyEdgeKernel.h
+++ b/arm_compute/core/NEON/kernels/NECannyEdgeKernel.h
@@ -86,7 +86,7 @@
};
#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
-/** NEON kernel to perform Gradient computation
+/** NEON kernel to perform Gradient computation for FP16 datatype
*/
class NEGradientFP16Kernel : public NEGradientKernel
{
@@ -99,6 +99,7 @@
void configure(const ITensor *gx, const ITensor *gy, ITensor *magnitude, ITensor *phase, int32_t norm_type) override;
};
#else /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */
+/** NEON kernel to perform Gradient computation for FP16 datatype */
using NEGradientFP16Kernel = NEGradientKernel;
#endif /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */
diff --git a/arm_compute/core/NEON/kernels/NEDepthwiseConvolutionLayer3x3Kernel.h b/arm_compute/core/NEON/kernels/NEDepthwiseConvolutionLayer3x3Kernel.h
index 5871cc5..0c2f30a 100644
--- a/arm_compute/core/NEON/kernels/NEDepthwiseConvolutionLayer3x3Kernel.h
+++ b/arm_compute/core/NEON/kernels/NEDepthwiseConvolutionLayer3x3Kernel.h
@@ -49,7 +49,7 @@
NEDepthwiseConvolutionLayer3x3Kernel &operator=(const NEDepthwiseConvolutionLayer3x3Kernel &) = delete;
/** Default Move Constructor. */
NEDepthwiseConvolutionLayer3x3Kernel(NEDepthwiseConvolutionLayer3x3Kernel &&) = default;
- /** Default move assignment operator. */
+ /** Default move assignment operator */
NEDepthwiseConvolutionLayer3x3Kernel &operator=(NEDepthwiseConvolutionLayer3x3Kernel &&) = default;
/** Initialize the function's source, destination, conv and border_size.
*
diff --git a/arm_compute/core/NEON/kernels/NEDequantizationLayerKernel.h b/arm_compute/core/NEON/kernels/NEDequantizationLayerKernel.h
index 7ee2078..f48e76f 100644
--- a/arm_compute/core/NEON/kernels/NEDequantizationLayerKernel.h
+++ b/arm_compute/core/NEON/kernels/NEDequantizationLayerKernel.h
@@ -50,7 +50,7 @@
NEDequantizationLayerKernel &operator=(const NEDequantizationLayerKernel &) = delete;
/** Default Move Constructor. */
NEDequantizationLayerKernel(NEDequantizationLayerKernel &&) = default;
- /** Default move assignment operator. */
+ /** Default move assignment operator */
NEDequantizationLayerKernel &operator=(NEDequantizationLayerKernel &&) = default;
/** Default destructor */
~NEDequantizationLayerKernel() = default;
diff --git a/arm_compute/core/NEON/kernels/NEGEMMMatrixVectorMultiplyKernel.h b/arm_compute/core/NEON/kernels/NEGEMMMatrixVectorMultiplyKernel.h
index 286be1a..a05d591 100644
--- a/arm_compute/core/NEON/kernels/NEGEMMMatrixVectorMultiplyKernel.h
+++ b/arm_compute/core/NEON/kernels/NEGEMMMatrixVectorMultiplyKernel.h
@@ -30,6 +30,7 @@
{
class ITensor;
+/** Interface for the GEMM matrix vector multiply kernel. **/
class NEGEMMMatrixVectorMultiplyKernel : public INESimpleKernel
{
public:
diff --git a/arm_compute/core/NEON/kernels/NEHarrisCornersKernel.h b/arm_compute/core/NEON/kernels/NEHarrisCornersKernel.h
index 8037e41..aabf8b3 100644
--- a/arm_compute/core/NEON/kernels/NEHarrisCornersKernel.h
+++ b/arm_compute/core/NEON/kernels/NEHarrisCornersKernel.h
@@ -127,6 +127,7 @@
HarrisScoreFunction *_func;
};
#else /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */
+/** Interface for the accumulate Weighted kernel using FP16 */
template <int32_t block_size>
using NEHarrisScoreFP16Kernel = NEHarrisScoreKernel<block_size>;
#endif /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */
diff --git a/arm_compute/core/NEON/kernels/NELKTrackerKernel.h b/arm_compute/core/NEON/kernels/NELKTrackerKernel.h
index f210558..9a8947f 100644
--- a/arm_compute/core/NEON/kernels/NELKTrackerKernel.h
+++ b/arm_compute/core/NEON/kernels/NELKTrackerKernel.h
@@ -45,6 +45,7 @@
bool tracking_status{ false }; /**< the tracking status of the keypoint */
};
+/** Interface for NEON Array of Internal Key Points. */
using INELKInternalKeypointArray = IArray<NELKInternalKeypoint>;
/** Interface for the Lucas-Kanade tracker kernel */
diff --git a/arm_compute/core/NEON/kernels/NEMagnitudePhaseKernel.h b/arm_compute/core/NEON/kernels/NEMagnitudePhaseKernel.h
index 522ed54..6967216 100644
--- a/arm_compute/core/NEON/kernels/NEMagnitudePhaseKernel.h
+++ b/arm_compute/core/NEON/kernels/NEMagnitudePhaseKernel.h
@@ -165,6 +165,7 @@
ITensor *_phase; /**< Output - Phase */
};
#else /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */
+/** Template interface for the kernel to compute magnitude and phase */
template <MagnitudeType mag_type, PhaseType phase_type>
using NEMagnitudePhaseFP16Kernel = NEMagnitudePhaseKernel<mag_type, phase_type>;
#endif /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */
diff --git a/arm_compute/core/NEON/kernels/NENonMaximaSuppression3x3Kernel.h b/arm_compute/core/NEON/kernels/NENonMaximaSuppression3x3Kernel.h
index f122ed1..588de49 100644
--- a/arm_compute/core/NEON/kernels/NENonMaximaSuppression3x3Kernel.h
+++ b/arm_compute/core/NEON/kernels/NENonMaximaSuppression3x3Kernel.h
@@ -83,7 +83,7 @@
};
#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
-/** NEON kernel to perform Non-Maxima suppression 3x3 with intermediate results in F16 if the input data type is F32
+/** NEON kernel to perform Non-Maxima suppression 3x3 with intermediate results in FP16 if the input data type is FP32
*/
class NENonMaximaSuppression3x3FP16Kernel : public NENonMaximaSuppression3x3Kernel
{
@@ -101,6 +101,7 @@
void configure(const ITensor *input, ITensor *output, bool border_undefined);
};
#else /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */
+/** NEON kernel to perform Non-Maxima suppression 3x3 with intermediate results in FP16 if the input data type is FP32 */
using NENonMaximaSuppression3x3FP16Kernel = NENonMaximaSuppression3x3Kernel;
#endif /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */
} // namespace arm_compute
diff --git a/arm_compute/core/NEON/kernels/NENormalizationLayerKernel.h b/arm_compute/core/NEON/kernels/NENormalizationLayerKernel.h
index b835ca7..6ae7b73 100644
--- a/arm_compute/core/NEON/kernels/NENormalizationLayerKernel.h
+++ b/arm_compute/core/NEON/kernels/NENormalizationLayerKernel.h
@@ -47,7 +47,7 @@
NENormalizationLayerKernel &operator=(const NENormalizationLayerKernel &) = delete;
/** Default Move Constructor. */
NENormalizationLayerKernel(NENormalizationLayerKernel &&) = default;
- /** Default move assignment operator. */
+ /** Default move assignment operator */
NENormalizationLayerKernel &operator=(NENormalizationLayerKernel &&) = default;
/** Default destructor */
~NENormalizationLayerKernel() = default;
diff --git a/arm_compute/core/NEON/kernels/NEQuantizationLayerKernel.h b/arm_compute/core/NEON/kernels/NEQuantizationLayerKernel.h
index e7cf0a8..ca7658b 100644
--- a/arm_compute/core/NEON/kernels/NEQuantizationLayerKernel.h
+++ b/arm_compute/core/NEON/kernels/NEQuantizationLayerKernel.h
@@ -50,7 +50,7 @@
NEQuantizationLayerKernel &operator=(const NEQuantizationLayerKernel &) = delete;
/** Default Move Constructor. */
NEQuantizationLayerKernel(NEQuantizationLayerKernel &&) = default;
- /** Default move assignment operator. */
+ /** Default move assignment operator */
NEQuantizationLayerKernel &operator=(NEQuantizationLayerKernel &&) = default;
/** Default destructor */
~NEQuantizationLayerKernel() = default;
diff --git a/arm_compute/core/NEON/kernels/NEWinogradLayerKernel.h b/arm_compute/core/NEON/kernels/NEWinogradLayerKernel.h
index 9169b75..2f44d19 100644
--- a/arm_compute/core/NEON/kernels/NEWinogradLayerKernel.h
+++ b/arm_compute/core/NEON/kernels/NEWinogradLayerKernel.h
@@ -34,6 +34,7 @@
{
class ITensor;
+/** Interface for the NEON kernel to perform Winograd input transform. */
template <typename T>
class INEWinogradLayerTransformInputKernel : public INEKernel
{
@@ -46,6 +47,8 @@
* @param[in] n_rows Number of rows in each feature map.
* @param[in] n_cols Number of columns in each feature map.
* @param[in] same_padding Use "SAME" padding, otherwise use "VALID".
+ *
+ * @return Storage size (in units of TIn) required.
*/
virtual unsigned int get_input_storage_size(int n_batches, int n_channels, int n_rows, int n_cols, bool same_padding) const = 0;
@@ -72,11 +75,13 @@
*/
virtual void configure(const T *const input, const int n_batches, const int n_rows, const int n_cols, const int n_channels, const PaddingType padding, T *const output, const int matrix_stride) = 0;
+ /** Destructor */
virtual ~INEWinogradLayerTransformInputKernel()
{
}
};
+/** NEON kernel to perform Winograd input transform. */
template <typename T, int OutputTileRows, int OutputTileCols, int KernelRows, int KernelCols>
class NEWinogradLayerTransformInputKernel : public INEWinogradLayerTransformInputKernel<T>
{
@@ -89,6 +94,8 @@
* @param[in] n_rows Number of rows in each feature map.
* @param[in] n_cols Number of columns in each feature map.
* @param[in] same_padding Use "SAME" padding, otherwise use "VALID".
+ *
+ * @return Storage size (in units of TIn) required.
*/
unsigned int get_input_storage_size(
int n_batches,
@@ -107,6 +114,7 @@
*/
int get_matrix_stride(const KernelShape &kernel_shape, const Tensor4DShape &input_shape, const PaddingType padding_type) const override;
+ /** Default constructor */
NEWinogradLayerTransformInputKernel();
const char *name() const override
@@ -139,7 +147,9 @@
void run(const Window &window, const ThreadInfo &info) override;
bool is_parallelisable() const override;
+ /** Winograd base kernel */
using WinogradBase = winograd::WinogradGEMM<OutputTileRows, OutputTileCols, KernelCols, KernelCols>;
+ /** Winograd convolution kernel */
using WinogradConv = typename WinogradBase::template Convolution<T, T>;
private:
@@ -147,6 +157,7 @@
std::unique_ptr<InputTransform> _transform;
};
+/** Interface for the NEON kernel to perform Winograd output transform. */
template <typename T>
class INEWinogradLayerTransformOutputKernel : public INEKernel
{
@@ -159,6 +170,8 @@
* @param[in] n_cols Number of columns in each feature map of the input tensor.
* @param[in] n_output_channels Number of feature maps in the output tensor.
* @param[in] same_padding Use "SAME" padding, otherwise use "VALID".
+ *
+ * @return Storage size (in units of TOut) required.
*/
virtual unsigned int get_output_storage_size(int n_batches, int n_rows, int n_cols, int n_output_channels, bool same_padding) const = 0;
@@ -208,6 +221,7 @@
}
};
+/** NEON kernel to perform Winograd output transform. */
template <typename T, int OutputTileRows, int OutputTileCols, int KernelRows, int KernelCols>
class NEWinogradLayerTransformOutputKernel : public INEWinogradLayerTransformOutputKernel<T>
{
@@ -227,7 +241,7 @@
NEWinogradLayerTransformOutputKernel(NEWinogradLayerTransformOutputKernel &&) = default;
/** Allow instances of this class to be moved */
NEWinogradLayerTransformOutputKernel &operator=(NEWinogradLayerTransformOutputKernel &&) = default;
-
+ /** Default destructor */
~NEWinogradLayerTransformOutputKernel() = default;
// Inherited methods overridden:
@@ -239,6 +253,8 @@
* @param[in] n_cols Number of columns in each feature map of the input tensor.
* @param[in] n_output_channels Number of feature maps in the output tensor.
* @param[in] same_padding Use "SAME" padding, otherwise use "VALID".
+ *
+ * @return Storage size (in units of TOut) required.
*/
unsigned int get_output_storage_size(int n_batches, int n_rows, int n_cols, int n_output_channels, bool same_padding) const override;
@@ -301,6 +317,7 @@
int _n_channels;
};
+/** Interface for the NEON kernel to perform Winograd weights transform. */
template <typename T>
class INEWinogradLayerTransformWeightsKernel : public INEKernel
{
@@ -310,6 +327,8 @@
*
* @param[in] n_output_channels Number of output feature maps.
* @param[in] n_input_channels Number of input feature maps.
+ *
+ * @return Storage size (in units of T) required.
*/
virtual unsigned int get_weight_storage_size(int n_output_channels, int n_input_channels) const = 0;
/** Gets the stride between matrices in the kernel worspace
@@ -335,10 +354,12 @@
}
};
+/** NEON kernel to perform Winograd weights transform. */
template <typename T, int OutputTileRows, int OutputTileCols, int KernelRows, int KernelCols>
class NEWinogradLayerTransformWeightsKernel final : public INEWinogradLayerTransformWeightsKernel<T>
{
public:
+ /** Default constructor. */
NEWinogradLayerTransformWeightsKernel();
const char *name() const override
{
@@ -359,6 +380,7 @@
std::unique_ptr<WeightsTransform> _transform;
};
+/** Interface for the NEON kernel to perform Winograd. */
template <typename TIn, typename TOut>
class INEWinogradLayerBatchedGEMMKernel : public INEKernel
{
@@ -406,16 +428,17 @@
virtual int get_number_blocks() const = 0;
};
+/** NEON kernel to perform Winograd. */
template <typename TIn, typename TOut, int OutputTileRows, int OutputTileCols, int KernelRows, int KernelCols>
class NEWinogradLayerBatchedGEMMKernel : public INEWinogradLayerBatchedGEMMKernel<TIn, TOut>
{
public:
+ /** Winograd base kernel */
using WinogradBase = winograd::WinogradGEMM<OutputTileRows, OutputTileCols, KernelRows, KernelCols>;
+ /** Winograd convolution kernel */
using WinogradConv = typename WinogradBase::template Convolution<TIn, TOut>;
- using MultiGEMM = winograd::BatchedBlockedGemm<WinogradConv::M_BLOCK, WinogradConv::N_BLOCK, TIn, TOut>;
-
- static const int _output_tile_rows = OutputTileRows;
- static const int _output_tile_cols = OutputTileCols;
+ /** Winograd batched blocked GEMM operator */
+ using MultiGEMM = winograd::BatchedBlockedGemm<WinogradConv::M_BLOCK, WinogradConv::N_BLOCK, TIn, TOut>;
const char *name() const override
{
@@ -432,7 +455,7 @@
NEWinogradLayerBatchedGEMMKernel(NEWinogradLayerBatchedGEMMKernel &&) = default;
/** Allow instances of this class to be moved */
NEWinogradLayerBatchedGEMMKernel &operator=(NEWinogradLayerBatchedGEMMKernel &&) = default;
-
+ /** Default destructor. */
~NEWinogradLayerBatchedGEMMKernel() = default;
// Inherited methods overridden:
@@ -474,6 +497,8 @@
void run(const Window &window, const ThreadInfo &info) override;
private:
+ static const int _output_tile_rows = OutputTileRows;
+ static const int _output_tile_cols = OutputTileCols;
std::unique_ptr<MultiGEMM> _gemms;
};
diff --git a/arm_compute/core/NEON/kernels/detail/NEActivationFunctionDetail.h b/arm_compute/core/NEON/kernels/detail/NEActivationFunctionDetail.h
index e4d3f54..71d5a9e 100644
--- a/arm_compute/core/NEON/kernels/detail/NEActivationFunctionDetail.h
+++ b/arm_compute/core/NEON/kernels/detail/NEActivationFunctionDetail.h
@@ -30,17 +30,25 @@
{
namespace detail
{
-// Dummy activation object
/** Dummy activation object */
template <typename T, int S>
struct dummy
{
+ /** NEON vector type. */
using ExactType = typename wrapper::traits::neon_vector<T, S>::type;
+ /** Construct a dummy activation object.
+ *
+ * @param[in] act_info Activation layer information.
+ */
explicit dummy(ActivationLayerInfo act_info)
{
ARM_COMPUTE_UNUSED(act_info);
}
+ /** Run activation function.
+ *
+ * @param[in] vval Vector of values.
+ */
void operator()(ExactType &vval)
{
ARM_COMPUTE_UNUSED(vval);
@@ -50,62 +58,97 @@
template <typename T, int S>
struct relu
{
- using ExactType = typename wrapper::traits::neon_vector<T, S>::type;
+ /** NEON vector type. */
+ using ExactType = typename wrapper::traits::neon_vector<T, S>::type;
+ /** NEON vector tag type. */
using ExactTagType = typename wrapper::traits::neon_vector<T, S>::tag_type;
+ /** Construct a RELU activation object.
+ *
+ * @param[in] act_info Activation layer information.
+ */
explicit relu(ActivationLayerInfo act_info)
: vzero(wrapper::vdup_n(static_cast<T>(0.f), ExactTagType{}))
{
ARM_COMPUTE_UNUSED(act_info);
}
+ /** Run activation function.
+ *
+ * @param[in] vval Vector of values.
+ */
void operator()(ExactType &vval)
{
vval = wrapper::vmax(vzero, vval);
}
+ /** Vector of zeroes. */
const ExactType vzero;
};
/** Bounded RELU activation object */
template <typename T, int S>
struct brelu
{
- using ExactType = typename wrapper::traits::neon_vector<T, S>::type;
+ /** NEON vector type. */
+ using ExactType = typename wrapper::traits::neon_vector<T, S>::type;
+ /** NEON vector tag type. */
using ExactTagType = typename wrapper::traits::neon_vector<T, S>::tag_type;
+ /** Construct a bounded RELU activation object.
+ *
+ * @param[in] act_info Activation layer information.
+ */
explicit brelu(ActivationLayerInfo act_info)
: vzero(wrapper::vdup_n(static_cast<T>(0.f), ExactTagType{})),
valpha(wrapper::vdup_n(static_cast<T>(act_info.a()), ExactTagType{}))
{
}
+ /** Run activation function.
+ *
+ * @param[in] vval Vector of values.
+ */
void operator()(ExactType &vval)
{
vval = wrapper::vmin(valpha, wrapper::vmax(vzero, vval));
}
+ /** Vector of zeroes. */
const ExactType vzero;
+ /** Vector of alphas. */
const ExactType valpha;
};
/** Lower-Upper Bounded RELU activation object */
template <typename T, int S>
struct lubrelu
{
- using ExactType = typename wrapper::traits::neon_vector<T, S>::type;
+ /** NEON vector type. */
+ using ExactType = typename wrapper::traits::neon_vector<T, S>::type;
+ /** NEON vector tag type. */
using ExactTagType = typename wrapper::traits::neon_vector<T, S>::tag_type;
+ /** Construct a lower-upper bounded RELU activation object.
+ *
+ * @param[in] act_info Activation layer information.
+ */
explicit lubrelu(ActivationLayerInfo act_info)
: valpha(wrapper::vdup_n(static_cast<T>(act_info.a()), ExactTagType{})),
vbeta(wrapper::vdup_n(static_cast<T>(act_info.b()), ExactTagType{}))
{
}
+ /** Run activation function.
+ *
+ * @param[in] vval Vector of values.
+ */
void operator()(ExactType &vval)
{
vval = wrapper::vmin(valpha, wrapper::vmax(vbeta, vval));
}
+ /** Vector of alphas. */
const ExactType valpha;
+ /** Vector of betas. */
const ExactType vbeta;
};
} // namespace detail