kernels Directory Reference

Directories
directory	fused

Files
file	add_kernels_bf16.c [code]
	Element-wise addition kernels for BF16 tensors.
file	attention_decode_fused.c [code]
	Fused attention decode kernel (legacy v6/v6.5)
file	attention_flash_true.c [code]
	Flash-style attention (online softmax, causal, streaming)
file	attention_kernels.c [code]
	Attention score/softmax/output kernels with SIMD (SSE/AVX/AVX512)
file	axpy_kernels.c [code]
	AXPY kernels for FP32: y = y + alpha * x.
file	dequant_kernels.c [code]
	Dequantization kernels for GGML-compatible formats.
file	embedding_kernels.c [code]
	Token/position embedding lookup kernels.
file	embedding_kernels_bf16.c [code]
	Token/position embedding lookup kernels for BF16.
file	fp16_convert.c [code]
	FP32 <-> FP16 SIMD conversion utilities.
file	gelu_kernels.c [code]
	GELU activation kernels with SIMD (SSE/AVX/AVX512)
file	gelu_kernels_bf16.c [code]
	GELU activation kernels for BF16 tensors.
file	gemm_batch_int8.c [code]
	Batch GEMM kernels for quantized weights with INT8 activations.
file	gemm_fused_kernels.c [code]
	Fused GEMM Kernels with activations.
file	gemm_head_major_output.c [code]
	Output projection from head-major attention (NO LAYOUT CONVERSION)
file	gemm_kernels.c [code]
	General matrix multiply (GEMM) kernels with SIMD (SSE/AVX/AVX512)
file	gemm_kernels_amx.c [code]
	AMX (Advanced Matrix Extensions) GEMM kernels.
file	gemm_kernels_bf16.c [code]
	Optimized BF16 GEMM Kernels for AVX-512.
file	gemm_kernels_f16.c [code]
	GEMM kernels with FP16 (half-precision) weights.
file	gemm_kernels_q4_0.c [code]
	GEMM/GEMV kernels with Q4_0 quantized weights.
file	gemm_kernels_q4_1.c [code]
	GEMM/GEMV kernels with Q4_1 quantized weights.
file	gemm_kernels_q4k.c [code]
	GEMM/GEMV kernels with Q4_K quantized weights.
file	gemm_kernels_q4k_avx.c [code]
	AVX Q4_K x Q8_K matvec kernel for Sandy/Ivy Bridge.
file	gemm_kernels_q4k_q8k.c [code]
	Q4_K (weights) x Q8_K (activations) kernels for inference.
file	gemm_kernels_q4k_q8k_avx2.c [code]
	AVX2 Q4_K x Q8_K matvec kernel (inference only)
file	gemm_kernels_q4k_q8k_vnni.c [code]
	VNNI Q4_K x Q8_K matvec kernel (inference only)
file	gemm_kernels_q4k_sse.c [code]
	SSE4.1 Q4_K x Q8_K dot product kernels.
file	gemm_kernels_q5_0.c [code]
	GEMM/GEMV kernels with Q5_0 quantized weights.
file	gemm_kernels_q5_0_sse.c [code]
	SSE4.1 GEMM for Q5_0 quantized weights.
file	gemm_kernels_q5_0_sse_v2.c [code]
	SSE-optimized GEMM kernels for Q5_0 x Q8_K quantization.
file	gemm_kernels_q5_1.c [code]
	GEMM/GEMV kernels with Q5_1 quantized weights.
file	gemm_kernels_q5_k.c [code]
	GEMM/GEMV kernels with Q5_K quantized weights.
file	gemm_kernels_q6k.c [code]
	GEMM/GEMV kernels with Q6_K quantized weights.
file	gemm_kernels_q6k_q8k.c [code]
	Q6_K (weights) x Q8_K (activations) kernels for inference.
file	gemm_kernels_q6k_sse.c [code]
	SSE-optimized GEMM kernels for Q6_K x Q8_K quantization.
file	gemm_kernels_q8_0.c [code]
	GEMM/GEMV kernels with Q8_0 quantized weights.
file	gemm_microkernel.c [code]
	GEMM Microkernel - High-Performance Register-Blocked Matrix Multiplication.
file	gemv_omp.c [code]
file	gemv_omp.h [code]
file	kv_cache_kernels.c [code]
	KV-cache helper kernels (head-major layout)
file	layernorm_kernels.c [code]
	LayerNorm forward/backward kernels with SIMD (SSE/AVX/AVX512)
file	layernorm_kernels_bf16.c [code]
	LayerNorm kernels for BF16 tensors.
file	loss_kernels.c [code]
	Loss function kernels (cross-entropy, etc.)
file	loss_kernels_bf16.c [code]
	Loss function kernels for BF16 tensors.
file	mlp_fused_decode.c [code]
	Fully fused MLP decode kernel (T=1 token generation)
file	mlp_kernels.c [code]
	MLP (feed-forward) kernels with SIMD (SSE/AVX/AVX512)
file	mlp_kernels_bf16.c [code]
	Optimized BF16 MLP Kernels.
file	optimizer_kernels.c [code]
	Optimizer kernels for training (AdamW, SGD)
file	optimizer_kernels_bf16.c [code]
	BF16 optimizer kernels for training.
file	qk_norm_kernels.c [code]
	Per-head RMSNorm on Q and K (Qwen3-style QK norm)
file	quantize_row_q8_k_sse.c [code]
	SSE-optimized Q8_K row quantization kernel.
file	relu_kernels.c [code]
	ReLU activation kernels with SIMD (SSE/AVX/AVX512)
file	relu_kernels_bf16.c [code]
	ReLU activation kernels for BF16 tensors.
file	rmsnorm_kernels.c [code]
	RMSNorm forward/backward kernels with SIMD (SSE/AVX/AVX512)
file	rmsnorm_kernels_bf16.c [code]
	RMSNorm kernels for BF16 tensors.
file	rmsnorm_kernels_int4.c [code]
	RMSNorm kernels with INT4 output quantization.
file	rmsnorm_kernels_int8.c [code]
	RMSNorm kernels with INT8 output quantization.
file	rope_kernels.c [code]
	RoPE (Rotary Position Embedding) kernels with SIMD.
file	rope_kernels_bf16.c [code]
	RoPE (Rotary Position Embedding) kernels for BF16.
file	sigmoid_kernels.c [code]
	Sigmoid activation kernels with SIMD (AVX512)
file	sigmoid_kernels_bf16.c [code]
	Sigmoid activation kernels for BF16 tensors.
file	softmax_kernels.c [code]
	Softmax forward/backward kernels with SIMD (SSE/AVX/AVX512)
file	softmax_kernels_bf16.c [code]
	Softmax kernels for BF16 tensors.
file	swiglu_kernels.c [code]
	SwiGLU activation kernels with SIMD (SSE/AVX/AVX512)
file	swiglu_kernels_bf16.c [code]
	SwiGLU activation kernels for BF16 tensors.
file	topk_kernels.c [code]
	Top-K selection kernels for MoE router dispatch.
file	vision_kernels.c [code]
	Vision kernels (im2patch, patch embedding, etc.)
file	vision_kernels_bf16.c [code]
	Vision kernels for BF16 tensors (im2patch, etc.)