diff options
author | Georgios Pinitas <georgios.pinitas@arm.com> | 2021-07-16 16:16:43 +0100 |
---|---|---|
committer | Georgios Pinitas <georgios.pinitas@arm.com> | 2021-07-22 02:25:50 +0000 |
commit | 4ee8b1599dbaf7634d25607fa5ac96ba3dc6b0f2 (patch) | |
tree | 2f8362d33cdad4212f4b96995681c68184c759e1 /src/core/NEON/kernels/arm_gemm/kernels/sve_hybrid_u8qa_mmla_4x4VL.hpp | |
parent | 59fd7a722e5bc7e85309d6200bc37a772721a719 (diff) | |
download | ComputeLibrary-4ee8b1599dbaf7634d25607fa5ac96ba3dc6b0f2.tar.gz |
Update GEMM assembly kernels
- Introduce Fp32 kernels with internal calculations in Bfloat16 when
fast_mode is enabled
- Improve kernel selection heuristics
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Change-Id: I68a9e7e862b6fd2721b46e0d7cc791091c4ab279
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5965
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/kernels/sve_hybrid_u8qa_mmla_4x4VL.hpp')
-rw-r--r-- | src/core/NEON/kernels/arm_gemm/kernels/sve_hybrid_u8qa_mmla_4x4VL.hpp | 101 |
1 files changed, 101 insertions, 0 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/kernels/sve_hybrid_u8qa_mmla_4x4VL.hpp b/src/core/NEON/kernels/arm_gemm/kernels/sve_hybrid_u8qa_mmla_4x4VL.hpp new file mode 100644 index 0000000000..da27554a0f --- /dev/null +++ b/src/core/NEON/kernels/arm_gemm/kernels/sve_hybrid_u8qa_mmla_4x4VL.hpp @@ -0,0 +1,101 @@ +/* + * Copyright (c) 2021 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 ARM_COMPUTE_ENABLE_SVE +#include "../std_transforms_sve.hpp" +#include "../performance_parameters.hpp" + +#define ARGLIST \ + unsigned int, const unsigned int *, \ + IndirectInputArg<uint8_t>, \ + size_t, size_t, \ + const uint8_t *, \ + IndirectOutputArg<uint8_t>, \ + const Requantize32 *, const int32_t *, unsigned int + +namespace arm_gemm +{ +// Actual kernel implementations +void sve_hybrid_u8qa_mmla_4x4VL( ARGLIST ); + +class cls_sve_hybrid_u8qa_mmla_4x4VL +{ +public: + typedef uint8_t lhs_operand_type; + typedef uint8_t rhs_operand_type; + typedef uint8_t result_type; + + typedef void (*kern_type)( ARGLIST ); + + /* Kernel blocking parameters */ + static constexpr unsigned int out_height() + { + return 4; + } + + static unsigned int out_width() + { + return get_vector_length<uint32_t>() * 4; + } + + static constexpr unsigned int k_unroll() + { + return 8; + } + + static constexpr bool supports_accumulate() + { + return false; + } + + StdTransformsSVE<rhs_operand_type, result_type, 4, 8, 8> transforms = {}; + template<typename T> + static inline PerformanceParameters get_performance_parameters(const CPUInfo *ci) + { + + if (std::is_same<T, uint8_t>::value) { + switch (ci->get_cpu_model()) { + default: + return { 47.30 }; + case CPUModel::A510: + return { 20.91 }; + } + } + + return { 1.0 }; + } + + // Default to the generic kernel + kern_type kernel=sve_hybrid_u8qa_mmla_4x4VL; + cls_sve_hybrid_u8qa_mmla_4x4VL(const CPUInfo *) + { + } +}; + +} // namespace arm_gemm + +#undef ARGLIST + +#endif // ARM_COMPUTE_ENABLE_SVE |