From ac4e873dad6aa6291fc36aff62047a896db04f6a Mon Sep 17 00:00:00 2001 From: Georgios Pinitas Date: Wed, 5 Jul 2017 17:02:25 +0100 Subject: COMPMID-417: Port DepthConcatenate to QS8/QS16 for NEON/CL. Change-Id: I3dddae63043c7aa18d908a4fc8abacf3c64f98ca Reviewed-on: http://mpd-gerrit.cambridge.arm.com/80081 Tested-by: Kaizen Reviewed-by: Steven Niu --- arm_compute/core/CL/CLHelpers.h | 8 ++++++++ arm_compute/core/CL/kernels/CLDepthConcatenateKernel.h | 4 ++-- 2 files changed, 10 insertions(+), 2 deletions(-) (limited to 'arm_compute/core/CL') diff --git a/arm_compute/core/CL/CLHelpers.h b/arm_compute/core/CL/CLHelpers.h index 01980d9793..eeb3e7699d 100644 --- a/arm_compute/core/CL/CLHelpers.h +++ b/arm_compute/core/CL/CLHelpers.h @@ -53,6 +53,14 @@ static constexpr const unsigned int max_cl_vector_width = 16; */ std::string get_cl_type_from_data_type(const DataType &dt); +/** Translates fixed point tensor data type to the underlying OpenCL type. + * + * @param[in] dt @ref DataType to be translated to OpenCL type. + * + * @return The string specifying the underlying OpenCL type to be used. + */ +std::string get_underlying_cl_type_from_data_type(const DataType &dt); + /** Translates a given gpu device target to string. * * @param[in] target Given gpu target. diff --git a/arm_compute/core/CL/kernels/CLDepthConcatenateKernel.h b/arm_compute/core/CL/kernels/CLDepthConcatenateKernel.h index eda4c66883..e85e0ec232 100644 --- a/arm_compute/core/CL/kernels/CLDepthConcatenateKernel.h +++ b/arm_compute/core/CL/kernels/CLDepthConcatenateKernel.h @@ -52,9 +52,9 @@ public: ~CLDepthConcatenateKernel() = default; /** Initialise the kernel's inputs and output * - * @param[in] input Input tensor. Data types supported: F32. + * @param[in] input Input tensor. Data types supported: QS8/QS16/F16/F32. * @param[in] depth_offset The offset on the Z axis. - * @param[in,out] output Output tensor. Data types supported: F32. + * @param[in,out] output Output tensor. Data types supported: Same as @p input. * * @note: The output tensor's low two dimensions can't be smaller than the input one's. * @note: The gaps between the two lowest dimensions of input and output need to be divisible by 2. -- cgit v1.2.1