Compute Library
 21.08
NESymm.h File Reference

Go to the source code of this file.

Namespaces

 arm_compute
 Copyright (c) 2017-2021 Arm Limited.
 

Typedefs

using qsymm8_t = int8_t
 8 bit quantized symmetric scalar value More...
 
using qsymm16x8_t = int16x8_t
 16 bit quantized symmetric vector with 8 elements More...
 
using qsymm16x8x2_t = int16x8x2_t
 16 bit quantized symmetric vector with 16 elements More...
 

Functions

template<bool is_bounded_relu>
int16x8_t finalize_quantization_int16 (int32x4x2_t &in_s32, int result_fixedpoint_multiplier, int32_t result_shift, int16x8_t min_s16, int16x8_t max_s16)
 Performs final quantization step on 8 signed 16-bit elements. More...
 
template<bool is_bounded_relu>
int16_t finalize_quantization_int16 (int32_t in_value, int result_fixedpoint_multiplier, int32_t result_shift, int16_t min_s16, int16_t max_s16)
 Performs final quantization step on single signed 16-bit element. More...
 
float32x4x2_t vdequantize_int16 (const int16x8_t &qv, float scale)
 Dequantize a neon vector holding 8 16-bit quantized values. More...
 
int16x8_t vquantize_int16 (const float32x4x2_t &qv, float scale)
 Quantize a neon vector holding 8 floating point values. More...
 
float32x4x4_t vdequantize (const int16x8x2_t &qv, const UniformQuantizationInfo &qi)
 Dequantize a neon vector holding 16 16-bit quantized values. More...
 
qsymm16x8x2_t vquantize_qsymm16 (const float32x4x4_t &qv, const UniformQuantizationInfo &qi)
 Quantize a neon vector holding 16 floating point values. More...
 
int32x4x2_t multiply_by_quantized_multiplier_2row (int32x4x2_t input, int32_t qmul, int32_t shift)
 Multiply a neon vector using quantized multiplier and shift. More...