From 7cd26d4a1b14bc4bf7c61496803416ab3d84791f Mon Sep 17 00:00:00 2001 From: Georgios Pinitas Date: Wed, 9 Jan 2019 18:35:17 +0000 Subject: COMPMID-1867: Add NEON/SVE GEMM Hybrid kernels. Change-Id: Ib40a9921e7f9a6a8be6c38872d6b3a0f24ed0cd3 Reviewed-on: https://review.mlplatform.org/515 Reviewed-by: Anthony Barbier Tested-by: Arm Jenkins --- .../a64_sgemm_nativeA_pretransposeB_16x4.hpp | 78 ++++++++++++++++++++++ 1 file changed, 78 insertions(+) create mode 100644 src/core/NEON/kernels/arm_gemm/kernels/a64_sgemm_nativeA_pretransposeB_16x4.hpp (limited to 'src/core/NEON/kernels/arm_gemm/kernels/a64_sgemm_nativeA_pretransposeB_16x4.hpp') diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_sgemm_nativeA_pretransposeB_16x4.hpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_sgemm_nativeA_pretransposeB_16x4.hpp new file mode 100644 index 0000000000..0c387ff6df --- /dev/null +++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_sgemm_nativeA_pretransposeB_16x4.hpp @@ -0,0 +1,78 @@ +/* + * Copyright (c) 2019 ARM Limited. + * + * SPDX-License-Identifier: MIT + * + * Permission is hereby granted, free of charge, to any person obtaining a copy + * of this software and associated documentation files (the "Software"), to + * deal in the Software without restriction, including without limitation the + * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or + * sell copies of the Software, and to permit persons to whom the Software is + * furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in all + * copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE + * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ +#pragma once + +#ifdef __aarch64__ + +namespace arm_gemm { + +// Actual kernel implementations +void a64_sgemm_nativeA_pretransposeB_16x4(const float *, int, const float *, float *, int, float, unsigned int, unsigned int, unsigned int); + +// Native A/Pretranspose B SGEMM "strategy" class. +// +// This describes the characteristics of a family of kernels, in terms of +// the required interleave properties and the output block size. +// +// All kernels in the family must share these characteristics. The actual +// kernel to be used can be chosen at runtime, based on the CPUInfo +// structure. +class sgemm_nativeA_pretransposeB_16x4 { +public: + typedef float operand_type; + typedef float result_type; + + typedef void (*kern_type)(const float *, int, const float *, float *, int, float, unsigned int, unsigned int, unsigned int); + + /* Desired data layout for B buffer (used for pretranspose) */ + static const int B_interleave = 16; + static const int B_block = 1; + static const bool B_transpose = true; + + /* Kernel blocking parameters */ + static int out_width() { + return 16; + } + + static int out_height() { + return 4; + } + + static int k_unroll() { + return 1; + } + + StdTransformsFixed transforms = {}; + + // Default to the generic kernel + kern_type kernel=a64_sgemm_nativeA_pretransposeB_16x4; + + sgemm_nativeA_pretransposeB_16x4(const CPUInfo *ci) { + + } +}; + +} // namespace arm_gemm + +#endif // __aarch64__ -- cgit v1.2.1