Module algorithms 

Core quantization algorithms Core quantization algorithms and tensor operations

This module provides the fundamental quantization and dequantization algorithms for tensor operations, including per-tensor and per-channel quantization schemes.

Features

• Per-tensor quantization: Single scale/zero-point for entire tensor
• Per-channel quantization: Individual scale/zero-point per channel
• Dequantization: Reverse quantization to restore floating point values
• Multiple schemes: Affine and symmetric quantization support
• Configuration-driven: Integration with QuantConfig for flexible usage

Structs

CacheAwareParams

Cache-aware quantization parameters for optimal memory access patterns

Functions

calculate_affine_quantization_params

Calculate affine quantization parameters (scale and zero_point)

calculate_symmetric_quantization_params

Calculate symmetric quantization parameters (scale only, zero_point = 0)

calculate_tensor_stats_cache_optimized

Cache-optimized tensor statistics calculation with blocking

dequantize

Dequantize a quantized tensor using scale and zero_point

dequantize_per_tensor_affine

Dequantize a tensor using per-tensor affine dequantization

get_cache_optimization_recommendations

Get cache-aware optimization recommendations for tensor operations

get_dtype_range

Get the quantization range for a given data type

quantize_auto

Convenience function to quantize with automatic parameter calculation

quantize_matrix_cache_friendly

Cache-friendly matrix quantization using tiling for 2D tensors

quantize_per_channel_auto

Auto-quantize a tensor using per-channel scheme

quantize_per_tensor

Quantize a tensor to INT8 using specified scale and zero point

quantize_per_tensor_affine

Quantize a tensor using per-tensor affine quantization

quantize_per_tensor_affine_cache_aware

Cache-aware per-tensor quantization optimized for memory hierarchy

quantize_per_tensor_affine_i8

Quantize a tensor using per-tensor affine quantization (returns I8 tensor)

quantize_streaming_with_prefetch

Prefetch-aware sequential quantization for streaming data

quantize_tensor_auto

Auto-quantize a tensor using per-tensor scheme

quantize_with_cache_optimization

Auto-select optimal quantization algorithm based on cache analysis

quantize_with_config

Quantize a tensor using specified configuration