diff options
author | Pablo Marquez Tello <pablo.tello@arm.com> | 2022-05-26 14:19:39 +0100 |
---|---|---|
committer | Pablo Marquez Tello <pablo.tello@arm.com> | 2022-06-07 12:33:48 +0000 |
commit | d75cd8ac5de1a785712e2e23c735a0167d64dfa3 (patch) | |
tree | 4e07c3cb0b159e14596c5e47d677de5877ec6efc /src/cpu/kernels/CpuActivationKernel.cpp | |
parent | fe1b1f6d94c196f086122613277ff95062a7e834 (diff) | |
download | ComputeLibrary-d75cd8ac5de1a785712e2e23c735a0167d64dfa3.tar.gz |
Compute Hard-Swish with a Lookup table for qasymm8.
* Resolves COMPMID-5211
Change-Id: I5914f971d733174dae67e6b4c589f45c75733cf7
Signed-off-by: Pablo Marquez Tello <pablo.tello@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/7654
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michalis Spyrou <michalis.spyrou@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/cpu/kernels/CpuActivationKernel.cpp')
-rw-r--r-- | src/cpu/kernels/CpuActivationKernel.cpp | 36 |
1 files changed, 23 insertions, 13 deletions
diff --git a/src/cpu/kernels/CpuActivationKernel.cpp b/src/cpu/kernels/CpuActivationKernel.cpp index 74148071ae..7d13817eab 100644 --- a/src/cpu/kernels/CpuActivationKernel.cpp +++ b/src/cpu/kernels/CpuActivationKernel.cpp @@ -47,52 +47,57 @@ static const std::vector<CpuActivationKernel::ActivationKernel> available_kernel { { "sve2_qu8_activation", - [](const DataTypeISASelectorData & data) { return data.dt == DataType::QASYMM8 && data.isa.sve2; }, + [](const ActivationDataTypeISASelectorData & data) { return data.dt == DataType::QASYMM8 && data.isa.sve2; }, REGISTER_QASYMM8_SVE2(arm_compute::cpu::sve2_qasymm8_activation) }, { "sve2_qs8_activation", - [](const DataTypeISASelectorData & data) { return data.dt == DataType::QASYMM8_SIGNED && data.isa.sve2; }, + [](const ActivationDataTypeISASelectorData & data) { return data.dt == DataType::QASYMM8_SIGNED && data.isa.sve2; }, REGISTER_QASYMM8_SIGNED_SVE2(arm_compute::cpu::sve2_qasymm8_signed_activation) }, { "sve2_qs16_activation", - [](const DataTypeISASelectorData & data) { return data.dt == DataType::QSYMM16 && data.isa.sve2; }, + [](const ActivationDataTypeISASelectorData & data) { return data.dt == DataType::QSYMM16 && data.isa.sve2; }, REGISTER_QSYMM16_SVE2(arm_compute::cpu::sve2_qsymm16_activation) }, { "sve_fp16_activation", - [](const DataTypeISASelectorData & data) { return data.dt == DataType::F16 && data.isa.sve && data.isa.fp16; }, + [](const ActivationDataTypeISASelectorData & data) { return data.dt == DataType::F16 && data.isa.sve && data.isa.fp16; }, REGISTER_FP16_SVE(arm_compute::cpu::sve_fp16_activation) }, { "sve_fp32_activation", - [](const DataTypeISASelectorData & data) { return data.dt == DataType::F32 && data.isa.sve; }, + [](const ActivationDataTypeISASelectorData & data) { return data.dt == DataType::F32 && data.isa.sve; }, REGISTER_FP32_SVE(arm_compute::cpu::sve_fp32_activation) }, { "neon_fp16_activation", - [](const DataTypeISASelectorData & data) { return data.dt == DataType::F16 && data.isa.fp16; }, + [](const ActivationDataTypeISASelectorData & data) { return data.dt == DataType::F16 && data.isa.fp16; }, REGISTER_FP16_NEON(arm_compute::cpu::neon_fp16_activation) }, { "neon_fp32_activation", - [](const DataTypeISASelectorData & data) { return data.dt == DataType::F32; }, + [](const ActivationDataTypeISASelectorData & data) { return data.dt == DataType::F32; }, REGISTER_FP32_NEON(arm_compute::cpu::neon_fp32_activation) }, { "neon_qu8_activation", - [](const DataTypeISASelectorData & data) { return data.dt == DataType::QASYMM8; }, + [](const ActivationDataTypeISASelectorData & data) { return data.dt == DataType::QASYMM8 && data.f != ActivationLayerInfo::ActivationFunction::HARD_SWISH; }, REGISTER_QASYMM8_NEON(arm_compute::cpu::neon_qasymm8_activation) }, { + "neon_qu8_activation_hardswish", + [](const ActivationDataTypeISASelectorData & data) { return data.dt == DataType::QASYMM8 && data.f == ActivationLayerInfo::ActivationFunction::HARD_SWISH; }, + REGISTER_QASYMM8_NEON(arm_compute::cpu::neon_qasymm8_hardswish_lut) + }, + { "neon_qs8_activation", - [](const DataTypeISASelectorData & data) { return data.dt == DataType::QASYMM8_SIGNED; }, + [](const ActivationDataTypeISASelectorData & data) { return data.dt == DataType::QASYMM8_SIGNED; }, REGISTER_QASYMM8_SIGNED_NEON(arm_compute::cpu::neon_qasymm8_signed_activation) }, { "neon_qs16_activation", - [](const DataTypeISASelectorData & data) { return data.dt == DataType::QSYMM16; }, + [](const ActivationDataTypeISASelectorData & data) { return data.dt == DataType::QSYMM16; }, REGISTER_QSYMM16_NEON(arm_compute::cpu::neon_qsymm16_activation) }, }; @@ -122,7 +127,7 @@ Status validate_arguments(const ITensorInfo *src, const ITensorInfo *dst, const ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(src); ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(src, 1, DataType::QASYMM8_SIGNED, DataType::QASYMM8, DataType::QSYMM16, DataType::F16, DataType::F32); - const auto *uk = CpuActivationKernel::get_implementation(DataTypeISASelectorData{ src->data_type(), CPUInfo::get().get_isa() }); + const auto *uk = CpuActivationKernel::get_implementation(ActivationDataTypeISASelectorData{ src->data_type(), CPUInfo::get().get_isa(), activation_info.activation() }); ARM_COMPUTE_RETURN_ERROR_ON(uk == nullptr || uk->ukernel == nullptr); @@ -176,14 +181,19 @@ void CpuActivationKernel::configure(const ITensorInfo *src, ITensorInfo *dst, Ac ARM_COMPUTE_ERROR_ON_NULLPTR(src); ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(src, dst, activation_info)); - const auto uk = CpuActivationKernel::get_implementation(DataTypeISASelectorData{ src->data_type(), CPUInfo::get().get_isa() }); + const auto uk = CpuActivationKernel::get_implementation(ActivationDataTypeISASelectorData{ src->data_type(), CPUInfo::get().get_isa(), activation_info.activation() }); ARM_COMPUTE_ERROR_ON_NULLPTR(uk); - _act_info = activation_info; _run_method = uk->ukernel; _name = std::string("CpuActivationKernel").append("/").append(uk->name); + if(activation_info.activation() == ActivationLayerInfo::ActivationFunction::HARD_SWISH && src->data_type() == DataType::QASYMM8) + { + activation_info.init_lut(src->quantization_info().uniform(), dst->quantization_info().uniform()); + } + _act_info = activation_info; + // Configure kernel window auto win_config = validate_and_configure_window(src, dst); ARM_COMPUTE_ERROR_THROW_ON(win_config.first); |