diff options
Diffstat (limited to 'src/core/NEON')
-rw-r--r-- | src/core/NEON/kernels/NEElementwiseOperationKernel.cpp | 27 | ||||
-rw-r--r-- | src/core/NEON/wrapper/intrinsics/intrinsics.h | 1 | ||||
-rw-r--r-- | src/core/NEON/wrapper/svtraits.h | 70 |
3 files changed, 95 insertions, 3 deletions
diff --git a/src/core/NEON/kernels/NEElementwiseOperationKernel.cpp b/src/core/NEON/kernels/NEElementwiseOperationKernel.cpp index 412ae247cb..29ae9037af 100644 --- a/src/core/NEON/kernels/NEElementwiseOperationKernel.cpp +++ b/src/core/NEON/kernels/NEElementwiseOperationKernel.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2018-2020 Arm Limited. + * Copyright (c) 2018-2021 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -29,6 +29,7 @@ #include "src/core/NEON/NEAsymm.h" #include "src/core/NEON/NEFixedPoint.h" #include "src/core/NEON/wrapper/wrapper.h" +#include "src/core/SVE/kernels/elementwise/impl/elementwise_list.h" #include "src/core/helpers/AutoConfiguration.h" #include "src/core/helpers/WindowHelpers.h" @@ -1135,14 +1136,23 @@ configure_arithm_func(const ITensorInfo *input1, const ITensorInfo *input2, ITen { static std::map<std::string, NEElementwiseOperationKernel::ElementwiseFunction *> map_function = { +#if defined(__ARM_FEATURE_SVE) + { "op_F32_F32_F32", &arm_compute::cpu::sve::elementwise_arithmetic_op<op, float32_t> }, + { "op_S32_S32_S32", &arm_compute::cpu::sve::elementwise_arithmetic_op<op, int32_t> }, +#else /* defined(__ARM_FEATURE_SVE) */ { "op_F32_F32_F32", &elementwise_arithm_op<op, typename wrapper::traits::neon_vector<float, 4>> }, - { "op_S16_S16_S16", &elementwise_arithm_op<op, typename wrapper::traits::neon_vector<int16_t, 8>> }, { "op_S32_S32_S32", &elementwise_arithm_op<op, typename wrapper::traits::neon_vector<int32_t, 4>> }, +#endif /* defined(__ARM_FEATURE_SVE) */ + { "op_S16_S16_S16", &elementwise_arithm_op<op, typename wrapper::traits::neon_vector<int16_t, 8>> }, { "op_QASYMM8_QASYMM8_QASYMM8", &elementwise_arithm_op_quantized<op> }, { "op_QASYMM8_SIGNED_QASYMM8_SIGNED_QASYMM8_SIGNED", &elementwise_arithm_op_quantized_signed<op> } }; #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC +#if defined(__ARM_FEATURE_SVE) + map_function["op_F16_F16_F16"] = &arm_compute::cpu::sve::elementwise_arithmetic_op<op, float16_t>; +#else /* defined(__ARM_FEATURE_SVE) */ map_function["op_F16_F16_F16"] = &elementwise_arithm_op<op, typename wrapper::traits::neon_vector<float16_t, 8>>; +#endif /* defined(__ARM_FEATURE_SVE) */ #endif /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */ return configure_func(input1, input2, output, map_function); @@ -1154,15 +1164,26 @@ configure_comp_func(const ITensorInfo *input1, const ITensorInfo *input2, ITenso { static std::map<std::string, NEElementwiseOperationKernel::ElementwiseFunction *> map_function = { +#if defined(__ARM_FEATURE_SVE) + { "op_U8_U8_U8", &arm_compute::cpu::sve::elementwise_comparison_op<op, uint8_t> }, + { "op_F32_F32_U8", &arm_compute::cpu::sve::elementwise_comparison_op<op, float> }, + { "op_S16_S16_U8", &arm_compute::cpu::sve::elementwise_comparison_op<op, int16_t> }, + { "op_S32_S32_U8", &arm_compute::cpu::sve::elementwise_comparison_op<op, int32_t> }, +#else /* defined(__ARM_FEATURE_SVE) */ { "op_U8_U8_U8", &elementwise_comp_op_8<op, uint8_t, uint8x16_t> }, { "op_F32_F32_U8", &elementwise_comp_op_32<op, float, float32x4_t> }, { "op_S16_S16_U8", &elementwise_comp_op_16<op, int16_t, int16x8_t> }, { "op_S32_S32_U8", &elementwise_comp_op_32<op, int32_t, int32x4_t> }, +#endif /* defined(__ARM_FEATURE_SVE) */ { "op_QASYMM8_SIGNED_QASYMM8_SIGNED_U8", &elementwise_comp_op_quantized_signed<op> }, { "op_QASYMM8_QASYMM8_U8", &elementwise_comp_op_quantized<op> } }; #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC - map_function["op_F16_F16_U8"] = &elementwise_comp_op_16<op, float16_t, float16x8_t>; +#if defined(__ARM_FEATURE_SVE) + map_function["op_F16_F16_U8"] = &arm_compute::cpu::sve::elementwise_comparison_op<op, float16_t>; +#else /* defined(__ARM_FEATURE_SVE) */ + map_function["op_F16_F16_U8"] = &elementwise_comp_op_16<op, float16_t, float16x8_t>; +#endif /* defined(__ARM_FEATURE_SVE) */ #endif /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */ return configure_func(input1, input2, output, map_function); diff --git a/src/core/NEON/wrapper/intrinsics/intrinsics.h b/src/core/NEON/wrapper/intrinsics/intrinsics.h index 4c7b674e2e..871d9cc5ac 100644 --- a/src/core/NEON/wrapper/intrinsics/intrinsics.h +++ b/src/core/NEON/wrapper/intrinsics/intrinsics.h @@ -79,6 +79,7 @@ #include "src/core/NEON/wrapper/intrinsics/svdup_n.h" #include "src/core/NEON/wrapper/intrinsics/svexp.h" #include "src/core/NEON/wrapper/intrinsics/svlog.h" +#include "src/core/NEON/wrapper/intrinsics/svpow.h" #include "src/core/NEON/wrapper/intrinsics/svptrue.h" #include "src/core/NEON/wrapper/intrinsics/svqadd.h" #include "src/core/NEON/wrapper/intrinsics/svsin.h" diff --git a/src/core/NEON/wrapper/svtraits.h b/src/core/NEON/wrapper/svtraits.h new file mode 100644 index 0000000000..465983d16f --- /dev/null +++ b/src/core/NEON/wrapper/svtraits.h @@ -0,0 +1,70 @@ +/* + * 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. + */ +#ifndef SRC_CORE_NEON_WRAPPER_SVTRAITS_H +#define SRC_CORE_NEON_WRAPPER_SVTRAITS_H +#if defined(__ARM_FEATURE_SVE) +#include "src/core/NEON/SVEMath.h" +#include <arm_sve.h> + +namespace arm_compute +{ +namespace wrapper +{ +template <typename T> +struct sve_scalar; +template <typename T> +struct sve_vector; + +#define DEFINE_TYPES(stype) \ + template <> \ + struct sve_scalar<sv##stype> \ + { \ + using type = stype; \ + }; \ + template <> \ + struct sve_vector<stype> \ + { \ + using type = sv##stype; \ + }; + +DEFINE_TYPES(int8_t) +DEFINE_TYPES(uint8_t) +DEFINE_TYPES(int16_t) +DEFINE_TYPES(uint16_t) +DEFINE_TYPES(int32_t) +DEFINE_TYPES(uint32_t) +DEFINE_TYPES(int64_t) +DEFINE_TYPES(uint64_t) +DEFINE_TYPES(float16_t) +DEFINE_TYPES(float32_t) +DEFINE_TYPES(float64_t) +DEFINE_TYPES(bfloat16_t) + +#undef DEFINE_TYPES + +} // namespace wrapper +} // namespace arm_compute + +#endif /* defined(__ARM_FEATURE_SVE) */ +#endif /* #ifndef SRC_CORE_NEON_WRAPPER_SVTRAITS_H */ |