Module avx2 

AVX2 and AVX-512 SIMD Implementations

Provides x86_64-specific SIMD implementations for CNN operations. Includes AVX2 (8 floats) and AVX-512 (16 floats) variants.

Functions

batch_norm_avx2^⚠

AVX2 batch normalization

conv_3x3_avx2^⚠

AVX2 3x3 convolution (without FMA, 4x loop unrolling + multiple accumulators)

conv_3x3_avx2_fma^⚠

AVX2 3x3 convolution with FMA (4x loop unrolling + multiple accumulators)

depthwise_conv_3x3_avx2^⚠

AVX2 depthwise 3x3 convolution (kernel position unrolling)

dot_product_avx2^⚠

AVX2 dot product without FMA

dot_product_avx2_fma^⚠

AVX2 dot product with FMA

dot_product_avx512^⚠

AVX-512 dot product

global_avg_pool_avx2^⚠

AVX2 global average pooling

max_pool_2x2_avx2^⚠

AVX2 max pooling 2x2

relu6_avx2^⚠

AVX2 ReLU6 activation

relu_avx2^⚠

AVX2 ReLU activation