From a3b1b469276b10484cd45901ae3a4b48b506caa9 Mon Sep 17 00:00:00 2001 From: Georgios Pinitas Date: Thu, 16 Nov 2017 19:24:39 +0000 Subject: COMPMID-667: Add validation static method to NEON GEMMlowp Change-Id: I8a470cc1351593ad8eeaf4ec92e04865e83d4f3c Reviewed-on: http://mpd-gerrit.cambridge.arm.com/96147 Tested-by: Kaizen Reviewed-by: Anthony Barbier --- arm_compute/core/ITensorInfo.h | 4 +++- .../core/NEON/kernels/NEGEMMInterleave4x4Kernel.h | 12 ++++++++++-- .../NEON/kernels/NEGEMMInterleaveBlockedKernel.h | 11 +++++++++++ .../NEON/kernels/NEGEMMLowpMatrixMultiplyKernel.h | 11 ++++++++++- .../kernels/NEGEMMLowpOffsetContributionKernel.h | 11 +++++++++++ .../NEGEMMLowpQuantizeDownInt32ToUint8ScaleKernel.h | 11 +++++++++++ .../core/NEON/kernels/NEGEMMLowpReductionKernel.h | 20 ++++++++++++++++++++ .../core/NEON/kernels/NEGEMMTranspose1xWKernel.h | 10 +++++++++- .../NEON/kernels/arm64/NEGEMMLowpAArch64V8P4Kernel.h | 10 ++++++++++ arm_compute/core/SubTensorInfo.h | 3 ++- arm_compute/core/TensorInfo.h | 3 ++- .../NEON/functions/NEGEMMLowpMatrixMultiplyCore.h | 9 +++++++++ .../runtime/NEON/functions/NEGEMMLowpOutputStage.h | 11 +++++++++++ 13 files changed, 119 insertions(+), 7 deletions(-) (limited to 'arm_compute') diff --git a/arm_compute/core/ITensorInfo.h b/arm_compute/core/ITensorInfo.h index 1bc0a80bac..4f69442b48 100644 --- a/arm_compute/core/ITensorInfo.h +++ b/arm_compute/core/ITensorInfo.h @@ -201,8 +201,10 @@ public: /** Set the flag whether the tensor size can be changed. * * @param[in] is_resizable Flag that marks the tensor if it can be changed or not. + * + * @return Reference to this ITensorInfo object */ - virtual void set_is_resizable(bool is_resizable) = 0; + virtual ITensorInfo &set_is_resizable(bool is_resizable) = 0; /** Valid region of the tensor. All elements in the valid region have defined values, i.e. are not undefined. * * @return The valid region. diff --git a/arm_compute/core/NEON/kernels/NEGEMMInterleave4x4Kernel.h b/arm_compute/core/NEON/kernels/NEGEMMInterleave4x4Kernel.h index 1c0d85c27b..db719caccb 100644 --- a/arm_compute/core/NEON/kernels/NEGEMMInterleave4x4Kernel.h +++ b/arm_compute/core/NEON/kernels/NEGEMMInterleave4x4Kernel.h @@ -56,10 +56,18 @@ public: NEGEMMInterleave4x4Kernel(); /** Initialise the kernel's input and output. * - * @param[in] input Input tensor. Data types supported: U8/S8/QS8/QS16/U16/S16/F16/U32/S32/F32 + * @param[in] input Input tensor. Data types supported: U8/S8/QS8/QASYMM8/QS16/U16/S16/F16/U32/S32/F32 * @param[out] output Output tensor which stores the interleaved matrix. Data type 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 NEGEMMInterleave4x4Kernel + * + * @param[in] input Input tensor info. Data types supported: U8/S8/QS8/QASYMM8/QS16/U16/S16/F16/U32/S32/F32 + * @param[in] output Output tensor info which stores the interleaved matrix. Data type supported: same as @p input. + * + * @return an error status + */ + static Error validate(const ITensorInfo *input, const ITensorInfo *output); // Inherited methods overridden: void run(const Window &window, const ThreadInfo &info) override; @@ -67,7 +75,7 @@ public: private: /** Common signature for all the transpose functions * - * @param[in] input An input tensor. Data types supported: U8/S8/QS8/U16/S16/QS16/F16/U32/S32/F32 + * @param[in] input An input tensor. Data types supported: U8/S8/QS8/QASYMM8/U16/S16/QS16/F16/U32/S32/F32 * @param[out] output The output tensor. Data type supported: same as @p input * @param[in] window Region on which to execute the kernel. */ diff --git a/arm_compute/core/NEON/kernels/NEGEMMInterleaveBlockedKernel.h b/arm_compute/core/NEON/kernels/NEGEMMInterleaveBlockedKernel.h index cdeb11d606..1a5b0fb863 100644 --- a/arm_compute/core/NEON/kernels/NEGEMMInterleaveBlockedKernel.h +++ b/arm_compute/core/NEON/kernels/NEGEMMInterleaveBlockedKernel.h @@ -50,6 +50,17 @@ public: * @param[in] transpose True if transpose operation must be performed, false otherwise. */ void configure(const ITensor *input, ITensor *output, unsigned int block_height, unsigned int block_width, bool transpose); + /** Static function to check if given info will lead to a valid configuration of @ref NEGEMMInterleaveBlockedKernel + * + * @param[in] input Input tensor. Data types supported: U8 + * @param[in] output Output tensor which stores the interleaved matrix. Data type supported: same as @p input. + * @param[in] block_height The height of the blocks to be interleaved. + * @param[in] block_width The width of the blocks to be interleaved. + * @param[in] transpose True if transpose operation must be performed, false otherwise. + * + * @return an error status + */ + Error validate(const ITensorInfo *input, const ITensorInfo *output, unsigned int block_height, unsigned int block_width, bool transpose); // Inherited methods overridden: void run(const Window &window, const ThreadInfo &info) override; diff --git a/arm_compute/core/NEON/kernels/NEGEMMLowpMatrixMultiplyKernel.h b/arm_compute/core/NEON/kernels/NEGEMMLowpMatrixMultiplyKernel.h index e9bfe4ea07..d9986b6cdd 100644 --- a/arm_compute/core/NEON/kernels/NEGEMMLowpMatrixMultiplyKernel.h +++ b/arm_compute/core/NEON/kernels/NEGEMMLowpMatrixMultiplyKernel.h @@ -58,11 +58,20 @@ public: * The input matrices @p input0 and @p input1 must be the output of the kernels: @ref NEGEMMInterleave4x4Kernel and @ref NEGEMMTranspose1xWKernel. These two * kernels change the layout of the original matrices to be more cache-friendly. * - * @param[in] input0 Input tensor containing the interleaved Matrix A. Data type supported: ASYMM8 + * @param[in] input0 Input tensor containing the interleaved Matrix A. Data type supported: QASYMM8 * @param[in] input1 Input tensor containing the transposed1xW Matrix B. Data type supported: same as @p input0 * @param[out] output Output tensor to store the result of matrix multiplication. Data type supported: S32 */ void configure(const ITensor *input0, const ITensor *input1, ITensor *output); + /** Static function to check if given info will lead to a valid configuration of @ref NEGEMMLowpMatrixMultiplyKernel + * + * @param[in] input0 Input tensor info containing the interleaved Matrix A. Data type supported: QASYMM8 + * @param[in] input1 Input tensor info containing the transposed Matrix B. Data type supported: same as @p input0 + * @param[in] output Output tensor info to store the result of matrix multiplication. Data type supported: S32 + * + * @return an error status + */ + static Error validate(const ITensorInfo *input0, const ITensorInfo *input1, const ITensorInfo *output); // Inherited methods overridden: void run(const Window &window, const ThreadInfo &info) override; diff --git a/arm_compute/core/NEON/kernels/NEGEMMLowpOffsetContributionKernel.h b/arm_compute/core/NEON/kernels/NEGEMMLowpOffsetContributionKernel.h index 8c1bae9396..27cb3f2c1c 100644 --- a/arm_compute/core/NEON/kernels/NEGEMMLowpOffsetContributionKernel.h +++ b/arm_compute/core/NEON/kernels/NEGEMMLowpOffsetContributionKernel.h @@ -68,6 +68,17 @@ public: * @param[in] b_offset Offset to be added to each element of the matrix B. */ void configure(ITensor *mm_result, const ITensor *vector_sum_col, const ITensor *vector_sum_row, int32_t k, int32_t a_offset, int32_t b_offset); + /** Static function to check if given info will lead to a valid configuration of @ref NEGEMMLowpOffsetContributionKernel + * + * @param[in] mm_result Input tensor containing the result of @ref NEGEMMLowpMatrixMultiplyKernel. Data type supported: S32 + * @param[in] vector_sum_col Input row-vector of sums of all the entries in each column of matrix B. + * Note: vector_sum_col can be a nullptr in case a_offset = 0. Data type supported: same as @p mm_result + * @param[in] vector_sum_row Input row-vector of sums of all the entries in each row of matrix A. + * Note: vector_sum_row can be a nullptr in case b_offset = 0. Data type supported: same as @p mm_result + * @param[in] a_offset Offset to be added to each element of the matrix A. + * @param[in] b_offset Offset to be added to each element of the matrix B. + */ + static Error validate(const ITensorInfo *mm_result, const ITensorInfo *vector_sum_col, const ITensorInfo *vector_sum_row, int32_t a_offset, int32_t b_offset); // Inherited methods overridden: void run(const Window &window, const ThreadInfo &info) override; diff --git a/arm_compute/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ToUint8ScaleKernel.h b/arm_compute/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ToUint8ScaleKernel.h index 4ec0e9df93..654dee21af 100644 --- a/arm_compute/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ToUint8ScaleKernel.h +++ b/arm_compute/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ToUint8ScaleKernel.h @@ -70,6 +70,17 @@ public: * Along with @p min, this value can be used to implement "rectified linear unit" activation functions */ void configure(const ITensor *input, const ITensor *bias, ITensor *output, int result_offset, int result_mult_int, int result_shift, int min = 0, int max = 0); + /** Static function to check if given info will lead to a valid configuration of @ref NEGEMMLowpQuantizeDownInt32ToUint8ScaleKernel + * + * @param[in] input Input tensor. Data type supported: S32 + * @param[in] bias Biases tensor. Only shared biases supported and it can be a nullptr if the biases addition is not required. + * Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p input. + * @param[in] output Output tensor. Data type supported: Data type supported: QASYMM8 + * @param[in] min (Optional) Min value used to saturate down the output result before converting back to QASYMM8 + * @param[in] max (Optional) Max value used to saturate up the output result before converting back to QASYMM8, + * Along with @p min, this value can be used to implement "rectified linear unit" activation functions + */ + static Error validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min = 0, int max = 0); // Inherited methods overridden: void run(const Window &window, const ThreadInfo &info) override; diff --git a/arm_compute/core/NEON/kernels/NEGEMMLowpReductionKernel.h b/arm_compute/core/NEON/kernels/NEGEMMLowpReductionKernel.h index 6eee54a9f0..9ca5cdf828 100644 --- a/arm_compute/core/NEON/kernels/NEGEMMLowpReductionKernel.h +++ b/arm_compute/core/NEON/kernels/NEGEMMLowpReductionKernel.h @@ -77,6 +77,16 @@ public: * @param[in] is_interleaved4x4 True if the matrix A has been interleaved4x4 */ void configure(const ITensor *mtx_a, ITensor *vector_sum_row, int32_t num_mtx_a_cols, bool is_interleaved4x4) override; + /** Static function to check if given info will lead to a valid configuration of @ref NEGEMMLowpMatrixAReductionKernel + * + * @param[in] mtx_a Input tensor. Data type supported: QASYMM8 + * @param[in] vector_sum_row Output row-vector of sums of all the entries in each row of mtx_a. Data type supported: S32 + * @param[in] num_mtx_a_cols Number of matrix A columns + * @param[in] is_interleaved4x4 True if the matrix A has been interleaved4x4 + * + * @return an error status + */ + static Error validate(const ITensorInfo *mtx_a, const ITensorInfo *vector_sum_row, int32_t num_mtx_a_cols, bool is_interleaved4x4); // Inherited methods overridden: void run(const Window &window, const ThreadInfo &info) override; @@ -98,6 +108,16 @@ public: * @param[in] is_transposed1xW True if the input tensor is transposed 1xW */ void configure(const ITensor *mtx_b, ITensor *vector_sum_col, int32_t num_mtx_b_rows, bool is_transposed1xW) override; + /** Static function to check if given info will lead to a valid configuration of @ref NEGEMMLowpMatrixBReductionKernel + * + * @param[in] mtx_b Input tensor. Data type supported: Data type supported: QASYMM8 + * @param[in] vector_sum_col Output row-vector of sums of all the entries in each column of mtx_b. Data type supported: S32 + * @param[in] num_mtx_b_rows Number of matrix B rows + * @param[in] is_transposed1xW True if the input tensor is transposed 1xW + * + * @return an error status + */ + static Error validate(const ITensorInfo *mtx_b, const ITensorInfo *vector_sum_col, int32_t num_mtx_b_rows, bool is_transposed1xW); // Inherited methods overridden: void run(const Window &window, const ThreadInfo &info) override; diff --git a/arm_compute/core/NEON/kernels/NEGEMMTranspose1xWKernel.h b/arm_compute/core/NEON/kernels/NEGEMMTranspose1xWKernel.h index 4d0bb2a484..4436d1fdb0 100644 --- a/arm_compute/core/NEON/kernels/NEGEMMTranspose1xWKernel.h +++ b/arm_compute/core/NEON/kernels/NEGEMMTranspose1xWKernel.h @@ -70,10 +70,18 @@ class NEGEMMTranspose1xWKernel : public INESimpleKernel public: /** Initialise the kernel's input and output. * - * @param[in] input Input tensor. Data types supported: U8/S8/QS8/U16/S16/QS16/F16/U32/S32/F32 + * @param[in] input Input tensor. Data types supported: U8/S8/QS8/QASYMM8/U16/S16/QS16/F16/U32/S32/F32 * @param[out] output Output tensor. Data type 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 NEGEMMTranspose1xWKernel + * + * @param[in] input Input tensor info. Data types supported: U8/S8/QS8/QASYMM8/U16/S16/QS16/F16/U32/S32/F32 + * @param[in] output Output tensor info. Data type supported: same as @p input. + * + * @return an error status + */ + static Error validate(const ITensorInfo *input, const ITensorInfo *output); // Inherited methods overridden: void run(const Window &window, const ThreadInfo &info) override; diff --git a/arm_compute/core/NEON/kernels/arm64/NEGEMMLowpAArch64V8P4Kernel.h b/arm_compute/core/NEON/kernels/arm64/NEGEMMLowpAArch64V8P4Kernel.h index 26af626aaa..66684a1185 100644 --- a/arm_compute/core/NEON/kernels/arm64/NEGEMMLowpAArch64V8P4Kernel.h +++ b/arm_compute/core/NEON/kernels/arm64/NEGEMMLowpAArch64V8P4Kernel.h @@ -39,6 +39,16 @@ class NEGEMMLowpAArch64V8P4Kernel : public NEGEMMAssemblyBaseKernel public: // Inherited methods overridden: void run(const Window &window, const ThreadInfo &info) override; + /** Static function to check if given info will lead to a valid configuration of @ref NEGEMMAssemblyBaseKernel + * + * The computed function is C = a * AxB + b * C. + * + * @param[in] input0 Input tensor info containing the Matrix A. Data types supported: QASYMM8 + * @param[in] input1 Input tensor info containing the Matrix B. Data types supported: same as @p input0 + * @param[in] output Output tensor info to store the result of matrix multiplication. + * If @p beta is not zero the values are multiplied by @p beta before the result is accumulated. Otherwise the values are overwritten by the result. Data types supported: S32 + */ + static Error validate(const ITensorInfo *input0, const ITensorInfo *input1, const ITensorInfo *output); protected: void internal_configure(const ITensor *input0, const ITensor *input1, ITensor *output, ITensor *workspace, float alpha, float beta, bool transform_0, bool transform_1) override; diff --git a/arm_compute/core/SubTensorInfo.h b/arm_compute/core/SubTensorInfo.h index b8a36854dc..7c464c0b17 100644 --- a/arm_compute/core/SubTensorInfo.h +++ b/arm_compute/core/SubTensorInfo.h @@ -185,10 +185,11 @@ public: ARM_COMPUTE_ERROR_ON(_parent == nullptr); return _parent->is_resizable(); } - void set_is_resizable(bool is_resizable) override + ITensorInfo &set_is_resizable(bool is_resizable) override { ARM_COMPUTE_ERROR_ON(_parent == nullptr); _parent->set_is_resizable(is_resizable); + return *this; } ValidRegion valid_region() const override { diff --git a/arm_compute/core/TensorInfo.h b/arm_compute/core/TensorInfo.h index 5fd6c47818..80ef7f8d5a 100644 --- a/arm_compute/core/TensorInfo.h +++ b/arm_compute/core/TensorInfo.h @@ -280,9 +280,10 @@ public: { return _is_resizable; } - void set_is_resizable(bool is_resizable) override + ITensorInfo &set_is_resizable(bool is_resizable) override { _is_resizable = is_resizable; + return *this; } ValidRegion valid_region() const override { diff --git a/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h b/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h index 0c441df4b9..598756e435 100644 --- a/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h +++ b/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h @@ -71,6 +71,15 @@ public: * @param[out] output Output tensor. Data type supported: Data type supported: S32 */ void configure(const ITensor *a, const ITensor *b, ITensor *output); + /** Static function to check if given info will lead to a valid configuration of @ref NEGEMMLowpMatrixMultiplyCore + * + * @param[in] a First input tensor (Matrix A). Data type supported: QASYMM8. + * @param[in] b Second input tensor (Matrix B). Data type supported: same as @p a + * @param[out] output Output tensor. Data type supported: Data type supported: S32 + * + * @return an error status + */ + static Error validate(const ITensorInfo *a, const ITensorInfo *b, const ITensorInfo *output); // Inherited methods overridden: void run() override; diff --git a/arm_compute/runtime/NEON/functions/NEGEMMLowpOutputStage.h b/arm_compute/runtime/NEON/functions/NEGEMMLowpOutputStage.h index a3db23aaee..9270d5581f 100644 --- a/arm_compute/runtime/NEON/functions/NEGEMMLowpOutputStage.h +++ b/arm_compute/runtime/NEON/functions/NEGEMMLowpOutputStage.h @@ -73,6 +73,17 @@ public: * Along with @p min, this value can be used to implement "rectified linear unit" activation functions */ void configure(const ITensor *input, const ITensor *bias, ITensor *output, int result_offset, int result_mult_int, int result_shift, int min = 0, int max = 0); + /** Static function to check if given info will lead to a valid configuration of @ref NEGEMMLowpQuantizeDownInt32ToUint8Scale + * + * @param[in] input Input tensor. It is the output of @ref NEGEMMLowpMatrixMultiplyCore function. Data type supported: S32 + * @param[in] bias Biases tensor. Only shared biases supported and it can be a nullptr if the addition of biases is not required. + * Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p input. + * @param[in] output Output tensor. Data type supported: Data type supported: QASYMM8 + * @param[in] min (Optional) Min value used to saturate down the output result before converting back to QASYMM8 + * @param[in] max (Optional) Max value used to saturate up the output result before converting back to QASYMM8, + * Along with @p min, this value can be used to implement "rectified linear unit" activation functions + */ + static Error validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min = 0, int max = 0); }; } #endif /*__ARM_COMPUTE_NEGEMMLOWPOUTPUTSTAGE_H__ */ \ No newline at end of file -- cgit v1.2.1