24.02.1
|
GEMMLowp output stage info. More...
#include <GEMMInfo.h>
Data Fields | |
GEMMLowpOutputStageType | type {GEMMLowpOutputStageType::NONE} |
GEMMLowp output stage type. More... | |
int32_t | gemmlowp_offset {0} |
GEMMLowp output stage offset used for quantizing to QASYMM8. More... | |
int32_t | gemmlowp_multiplier {0} |
GEMMLowp output stage multiplier used for quantizing to QASYMM8. More... | |
int32_t | gemmlowp_shift {0} |
GEMMLowp output stage shift used for quantizing to uint8. More... | |
int32_t | gemmlowp_min_bound |
GEMMLowp min value used to saturate down the output result before converting back to QASYMM8. More... | |
int32_t | gemmlowp_max_bound |
GEMMLowp max value used to saturate down the output result before converting back to QASYMM8. More... | |
std::vector< int32_t > | gemmlowp_multipliers {} |
GEMMLowp output stage multiplier used for quantizing to QASYMM8. More... | |
std::vector< int32_t > | gemmlowp_shifts {} |
GEMMLowp output stage multiplier used for quantizing to QASYMM8. More... | |
float | gemmlowp_real_multiplier {0} |
GEMMLowp output stage real multiplier used for quantizing to QASYMM8. More... | |
bool | is_quantized_per_channel {false} |
GEMMLowp quantized per-channel flag. More... | |
DataType | output_data_type |
Output tensor data type to use if the output is not initialized. More... | |
GEMMLowp output stage info.
Definition at line 45 of file GEMMInfo.h.
int32_t gemmlowp_max_bound |
GEMMLowp max value used to saturate down the output result before converting back to QASYMM8.
Definition at line 54 of file GEMMInfo.h.
Referenced by CpuGemmLowpQuantizeDownInt32ScaleKernel::configure(), ClGemmConv2d::configure(), CpuGemmLowpMatrixMultiplyCore::configure(), NEQLSTMLayer::configure(), CLQLSTMLayer::configure(), arm_compute::operator<<(), CpuGemmLowpOffsetContributionOutputStageKernel::run_op(), ClGemmConv2d::validate(), CpuGemmLowpMatrixMultiplyCore::validate(), NEQLSTMLayer::validate(), and CLQLSTMLayer::validate().
int32_t gemmlowp_min_bound |
GEMMLowp min value used to saturate down the output result before converting back to QASYMM8.
Definition at line 51 of file GEMMInfo.h.
Referenced by CpuGemmLowpQuantizeDownInt32ScaleKernel::configure(), ClGemmConv2d::configure(), CpuGemmLowpMatrixMultiplyCore::configure(), NEQLSTMLayer::configure(), CLQLSTMLayer::configure(), arm_compute::operator<<(), CpuGemmLowpOffsetContributionOutputStageKernel::run_op(), ClGemmConv2d::validate(), CpuGemmLowpMatrixMultiplyCore::validate(), NEQLSTMLayer::validate(), and CLQLSTMLayer::validate().
int32_t gemmlowp_multiplier {0} |
GEMMLowp output stage multiplier used for quantizing to QASYMM8.
Definition at line 49 of file GEMMInfo.h.
Referenced by arm_compute::quantization::calculate_quantized_multipliers(), ClGemmConv2d::configure(), NEQLSTMLayer::configure(), CLQLSTMLayer::configure(), arm_compute::operator<<(), ClGemmConv2d::validate(), NEQLSTMLayer::validate(), and CLQLSTMLayer::validate().
std::vector<int32_t> gemmlowp_multipliers {} |
GEMMLowp output stage multiplier used for quantizing to QASYMM8.
Definition at line 57 of file GEMMInfo.h.
Referenced by arm_compute::quantization::calculate_quantized_multipliers(), ClGemmLowpMatrixMultiplyCore::configure(), ClGemmConv2d::configure(), ClGemmLowpMatrixMultiplyCore::prepare(), ClGemmLowpMatrixMultiplyCore::validate(), and ClGemmConv2d::validate().
int32_t gemmlowp_offset {0} |
GEMMLowp output stage offset used for quantizing to QASYMM8.
Definition at line 48 of file GEMMInfo.h.
Referenced by ClGemmConv2d::configure(), CpuGemmLowpMatrixMultiplyCore::configure(), NEQLSTMLayer::configure(), CLQLSTMLayer::configure(), arm_compute::operator<<(), ClGemmConv2d::validate(), CpuGemmLowpMatrixMultiplyCore::validate(), NEQLSTMLayer::validate(), and CLQLSTMLayer::validate().
float gemmlowp_real_multiplier {0} |
GEMMLowp output stage real multiplier used for quantizing to QASYMM8.
Definition at line 59 of file GEMMInfo.h.
Referenced by arm_compute::operator<<().
int32_t gemmlowp_shift {0} |
GEMMLowp output stage shift used for quantizing to uint8.
Definition at line 50 of file GEMMInfo.h.
Referenced by arm_compute::quantization::calculate_quantized_multipliers(), ClGemmConv2d::configure(), NEQLSTMLayer::configure(), CLQLSTMLayer::configure(), arm_compute::operator<<(), ClGemmConv2d::validate(), NEQLSTMLayer::validate(), and CLQLSTMLayer::validate().
std::vector<int32_t> gemmlowp_shifts {} |
GEMMLowp output stage multiplier used for quantizing to QASYMM8.
Definition at line 58 of file GEMMInfo.h.
Referenced by arm_compute::quantization::calculate_quantized_multipliers(), ClGemmConv2d::configure(), ClGemmLowpMatrixMultiplyCore::prepare(), and ClGemmConv2d::validate().
bool is_quantized_per_channel {false} |
GEMMLowp quantized per-channel flag.
Definition at line 60 of file GEMMInfo.h.
Referenced by ClGemmLowpMatrixMultiplyCore::configure(), ClGemmConv2d::configure(), arm_compute::operator<<(), ClGemmLowpMatrixMultiplyCore::prepare(), CpuGemmLowpOffsetContributionOutputStageKernel::run_op(), ClGemmLowpMatrixMultiplyCore::validate(), and ClGemmConv2d::validate().
DataType output_data_type |
Output tensor data type to use if the output is not initialized.
Definition at line 61 of file GEMMInfo.h.
Referenced by CpuGemmLowpQuantizeDownInt32ScaleKernel::configure(), ClGemmLowpMatrixMultiplyCore::configure(), NEQLSTMLayer::configure(), CLQLSTMLayer::configure(), arm_compute::operator<<(), ClGemmLowpMatrixMultiplyCore::validate(), NEQLSTMLayer::validate(), and CLQLSTMLayer::validate().
GEMMLowp output stage type.
Definition at line 47 of file GEMMInfo.h.
Referenced by ClGemmLowpMatrixMultiplyCore::configure(), ClGemmConv2d::configure(), NEQLSTMLayer::configure(), CLQLSTMLayer::configure(), arm_compute::operator<<(), CpuGemmLowpMatrixMultiplyCore::run(), CpuGemmLowpOffsetContributionOutputStageKernel::run_op(), ClGemmLowpMatrixMultiplyCore::validate(), ClGemmConv2d::validate(), CpuGemmLowpMatrixMultiplyCore::validate(), NEQLSTMLayer::validate(), and CLQLSTMLayer::validate().