From 9fb2c61b256ccf66b6a7b6f605008125288d60cf Mon Sep 17 00:00:00 2001 From: Cedric Nugteren Date: Sun, 7 Jan 2018 14:27:15 +0100 Subject: Added API and tests for new GemmStridedBatched routine --- include/clblast.h | 12 ++++++++++++ 1 file changed, 12 insertions(+) (limited to 'include/clblast.h') diff --git a/include/clblast.h b/include/clblast.h index a05b487f..8e3e64da 100644 --- a/include/clblast.h +++ b/include/clblast.h @@ -647,6 +647,18 @@ StatusCode GemmBatched(const Layout layout, const Transpose a_transpose, const T const size_t batch_count, cl_command_queue* queue, cl_event* event = nullptr); +// StridedBatched version of GEMM: SGEMMSTRIDEDBATCHED/DGEMMSTRIDEDBATCHED/CGEMMSTRIDEDBATCHED/ZGEMMSTRIDEDBATCHED/HGEMMSTRIDEDBATCHED +template +StatusCode GemmStridedBatched(const Layout layout, const Transpose a_transpose, const Transpose b_transpose, + const size_t m, const size_t n, const size_t k, + const T alpha, + const cl_mem a_buffer, const size_t a_offset, const size_t a_ld, const size_t a_stride, + const cl_mem b_buffer, const size_t b_offset, const size_t b_ld, const size_t b_stride, + const T beta, + cl_mem c_buffer, const size_t c_offset, const size_t c_ld, const size_t c_stride, + const size_t batch_count, + cl_command_queue* queue, cl_event* event = nullptr); + // ================================================================================================= // Retrieves the required size of the temporary buffer for the GEMM kernel (optional) -- cgit v1.2.3