diff options
Diffstat (limited to 'src/core/NEON/kernels/assembly')
| -rw-r--r-- | src/core/NEON/kernels/assembly/depthwise.hpp | 7 | ||||
| -rw-r--r-- | src/core/NEON/kernels/assembly/pool_common.hpp | 10 | ||||
| -rw-r--r-- | src/core/NEON/kernels/assembly/pooling.hpp | 13 |
3 files changed, 25 insertions, 5 deletions
diff --git a/src/core/NEON/kernels/assembly/depthwise.hpp b/src/core/NEON/kernels/assembly/depthwise.hpp index 13c2d314e4..0b68cb4db8 100644 --- a/src/core/NEON/kernels/assembly/depthwise.hpp +++ b/src/core/NEON/kernels/assembly/depthwise.hpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2021-2023 Arm Limited. + * Copyright (c) 2021-2024 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -22,6 +22,9 @@ * SOFTWARE. */ +#ifndef ACL_SRC_CORE_NEON_KERNELS_ASSEMBLY_DEPTHWISE_HPP +#define ACL_SRC_CORE_NEON_KERNELS_ASSEMBLY_DEPTHWISE_HPP + #pragma once #include "arm_gemm.hpp" @@ -349,3 +352,5 @@ std::vector<KernelDescription> get_compatible_kernels(const DepthwiseArgs &, con } // namespace depthwise } // namespace arm_conv + +#endif // ACL_SRC_CORE_NEON_KERNELS_ASSEMBLY_DEPTHWISE_HPP diff --git a/src/core/NEON/kernels/assembly/pool_common.hpp b/src/core/NEON/kernels/assembly/pool_common.hpp index 045f9f95d3..a2d87e9382 100644 --- a/src/core/NEON/kernels/assembly/pool_common.hpp +++ b/src/core/NEON/kernels/assembly/pool_common.hpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2021-2023 Arm Limited. + * Copyright (c) 2021-2024 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -22,6 +22,9 @@ * SOFTWARE. */ +#ifndef ACL_SRC_CORE_NEON_KERNELS_ASSEMBLY_POOL_COMMON_HPP +#define ACL_SRC_CORE_NEON_KERNELS_ASSEMBLY_POOL_COMMON_HPP + #pragma once #ifdef CYCLE_PROFILING #include "profiler.hpp" @@ -65,7 +68,8 @@ public: virtual ~IPoolingCommon() = default; // Determine the amount of working space required. - virtual size_t get_working_size(unsigned int num_threads) const = 0; + virtual size_t get_working_size(unsigned int num_threads) const = 0; + virtual size_t get_working_size(unsigned int num_threads, unsigned int n_channels) const = 0; // Execute pooling over the specified area of memory. virtual void execute(const void *const input, @@ -108,3 +112,5 @@ public: } // namespace pooling } // namespace arm_conv + +#endif // ACL_SRC_CORE_NEON_KERNELS_ASSEMBLY_POOL_COMMON_HPP diff --git a/src/core/NEON/kernels/assembly/pooling.hpp b/src/core/NEON/kernels/assembly/pooling.hpp index 89d594298e..d64a59f4d0 100644 --- a/src/core/NEON/kernels/assembly/pooling.hpp +++ b/src/core/NEON/kernels/assembly/pooling.hpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2021-2023 Arm Limited. + * Copyright (c) 2021-2024 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -22,6 +22,9 @@ * SOFTWARE. */ +#ifndef ACL_SRC_CORE_NEON_KERNELS_ASSEMBLY_POOLING_HPP +#define ACL_SRC_CORE_NEON_KERNELS_ASSEMBLY_POOLING_HPP + #pragma once #include "arm_gemm_local.hpp" @@ -136,7 +139,11 @@ public: PoolingCommon(PoolingCommon &) = delete; PoolingCommon &operator=(PoolingCommon &) = delete; - size_t get_working_size(unsigned int) const override = 0; + size_t get_working_size(unsigned int, unsigned int) const override = 0; + size_t get_working_size(unsigned int n_threads) const override + { + return this->get_working_size(n_threads, m_args.n_channels); + } // Execute pooling over the specified area of memory. void execute(const void *const input, @@ -223,3 +230,5 @@ UniquePoolingCommon<TInput, TOutput> pooling(const PoolingArgs &, const OutputSt } // namespace pooling } // namespace arm_conv + +#endif // ACL_SRC_CORE_NEON_KERNELS_ASSEMBLY_POOLING_HPP |