From 6ff12a0f7765f62b8d0fa8554021e1cac2789f19 Mon Sep 17 00:00:00 2001 From: Pablo Tello Date: Thu, 2 Nov 2017 16:09:35 +0000 Subject: COMPMID-662: Integrated the new a64_s8_gemm_12x8 + dot product kernel into ACL. Change-Id: Id8f919e486a132fc58346c9f84fccbeeb83d19b3 Reviewed-on: http://mpd-gerrit.cambridge.arm.com/94233 Tested-by: Kaizen Reviewed-by: Anthony Barbier Reviewed-by: Gian Marco Iodice --- arm_compute/runtime/NEON/functions/NEGEMMLowp.h | 8 ++++---- arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h | 1 + 2 files changed, 5 insertions(+), 4 deletions(-) (limited to 'arm_compute/runtime/NEON/functions') diff --git a/arm_compute/runtime/NEON/functions/NEGEMMLowp.h b/arm_compute/runtime/NEON/functions/NEGEMMLowp.h index 865c729d39..59c919e161 100644 --- a/arm_compute/runtime/NEON/functions/NEGEMMLowp.h +++ b/arm_compute/runtime/NEON/functions/NEGEMMLowp.h @@ -57,14 +57,14 @@ public: * @note GEMM_LOWP: low precision GEMM kernel * This kernel performs the following computations: * - * -# Convert a values from uint8 to int32 and add a_offset to each of them. - * -# Convert b values from uint8 to int32 and add b_offset to each of them. + * -# Convert a values from int8 to int32 and add a_offset to each of them. + * -# Convert b values from int8 to int32 and add b_offset to each of them. * -# Compute the int32 matrix product of the resulting a * b. * -# Add output_offset to each entry of the result. * -# Multiply each entry of the result and round to the nearest integer - * -# Clamp the resulting int32 values to the [0..255] range and cast to uint8. + * -# Clamp the resulting int32 values to the [0..255] range and cast to int8. * - * @param[in] a First input tensor (Matrix A). Data type supported: U8. + * @param[in] a First input tensor (Matrix A). Data type supported: S8. * @param[in] b Second input tensor (Matrix B). Data type supported: same as @p a * @param[out] output Output tensor. Data type supported: same as @p a. * @param[in] a_offset Offset to be added to each element of the matrix A. diff --git a/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h b/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h index 4866e781f5..c81a432295 100644 --- a/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h +++ b/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h @@ -78,6 +78,7 @@ private: std::unique_ptr _mtx_b_reshape_kernel; Tensor _tmp_a; Tensor _tmp_b; + Tensor _workspace; }; } #endif /*__ARM_COMPUTE_NEGEMMLOWPMATRIXMULTIPLYCORE_H__ */ -- cgit v1.2.1