diff options
author | Milos Puzovic <milos.puzovic@arm.com> | 2024-03-26 14:34:30 +0000 |
---|---|---|
committer | Milos Puzovic <milos.puzovic@arm.com> | 2024-03-27 10:27:49 +0000 |
commit | 905786ea0c1abb2b8df36c56eae93a97823cace1 (patch) | |
tree | 7578977256edcffcd08bf18f839ae647f9691179 /src/core/NEON/kernels/arm_gemm/kernels/a64_ffhybrid_fp32bf16fp32_mmla_6x16.hpp | |
parent | 37d84451eb1f8b4811faa4359ce154c832602782 (diff) | |
download | ComputeLibrary-905786ea0c1abb2b8df36c56eae93a97823cace1.tar.gz |
Added new NEON fixed format fast math mode hybrid kernel with maximum height of 6 for accumulation and updated heuristics
Change-Id: Ib52ea6825e164f4a8b8422eab7991b50af0b0d7c
Signed-off-by: Milos Puzovic <milos.puzovic@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/11354
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Jakub Sujak <jakub.sujak@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/kernels/a64_ffhybrid_fp32bf16fp32_mmla_6x16.hpp')
-rw-r--r-- | src/core/NEON/kernels/arm_gemm/kernels/a64_ffhybrid_fp32bf16fp32_mmla_6x16.hpp | 111 |
1 files changed, 111 insertions, 0 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_ffhybrid_fp32bf16fp32_mmla_6x16.hpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_ffhybrid_fp32bf16fp32_mmla_6x16.hpp new file mode 100644 index 0000000000..98f7fc9403 --- /dev/null +++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_ffhybrid_fp32bf16fp32_mmla_6x16.hpp @@ -0,0 +1,111 @@ +/* + * Copyright (c) 2024 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__ + +#include "../std_transforms_fixed.hpp" +#include "../bfloat.hpp" +#include "../kernel_weight_format.hpp" +#include "../performance_parameters.hpp" + +#define ARGLIST \ + unsigned int, const unsigned int *, \ + IndirectInputArg<float>, \ + size_t, size_t, \ + const bfloat16 *, \ + size_t, \ + IndirectOutputArg<float>, \ + const float *, Activation, bool + +namespace arm_gemm +{ +// Actual kernel implementations +void a64_ffhybrid_fp32bf16fp32_mmla_6x16( ARGLIST ); + +class cls_a64_ffhybrid_fp32bf16fp32_mmla_6x16 +{ +public: + typedef float lhs_operand_type; + typedef bfloat16 rhs_operand_type; + typedef float result_type; + + typedef void (*kern_type)( ARGLIST ); + + /* Kernel blocking parameters */ + static constexpr unsigned int out_height() + { + return 6; + } + static unsigned int stripe_width() + { + return 4; + } + + static KernelWeightFormat kernel_weight_format() + { + return KernelWeightFormat::VL256_BL64_BF16; + } + + static unsigned int out_width() + { + return 16; + } + + static constexpr unsigned int k_unroll() + { + return 4; + } + + static constexpr bool supports_accumulate() + { + return true; + } + + StdTransformsFixed<rhs_operand_type, result_type, 6, 16, 4> transforms = {}; + template<typename T> + static inline PerformanceParameters get_performance_parameters(const CPUInfo *ci) + { + if (std::is_same<T, float>::value) { + switch (ci->get_cpu_model()) { + case CPUModel::V1: + return { 21.05 }; + default: + return { 15.27 }; + } + } + + return { 1.0 }; + } + + // Default to the generic kernel + kern_type kernel=a64_ffhybrid_fp32bf16fp32_mmla_6x16; + cls_a64_ffhybrid_fp32bf16fp32_mmla_6x16(const CPUInfo *) + { + } +}; + +} // namespace arm_gemm + +#undef ARGLIST +#endif // __aarch64__ |