Questions regarding BitNet Integration in llama.cpp Inference Pipeline

I noticed that we’ve defined several custom functions for BitNet inference. However, I’m curious—how does llama.cpp know when and where to call these functions? Specifically, how are these BitNet-related functions integrated or injected into the inference flow of llama.cpp?

Especially all these functions.

GGML_API void ggml_bitnet_init(void);
GGML_API void ggml_bitnet_free(void);
// src0->type == Q4_0/IQ2_XXS/IQ3_XXS
// bitnet.cpp currently only supports BitNet quantization or GPTQ-like quantization (only scales, without zeros)
// If use i-quantization gguf models, the results will be wrong
// TODO: add customized block types Q2_0/Q3_0
GGML_API bool ggml_bitnet_can_mul_mat(const struct ggml_tensor * src0, const struct ggml_tensor * src1, const struct ggml_tensor * dst);
GGML_API size_t ggml_bitnet_mul_mat_get_wsize(const struct ggml_tensor * src0, const struct ggml_tensor * src1, const struct ggml_tensor * dst);
GGML_API void ggml_bitnet_mul_mat_task_init(void * src1, void * qlut, void * lut_scales, void * lut_biases, int n, int k, int m, int bits);
GGML_API void ggml_bitnet_mul_mat_task_compute(void * src0, void * scales, void * qlut, void * lut_scales, void * lut_biases, void * dst, int n, int k, int m, int bits);
GGML_API void ggml_bitnet_transform_tensor(struct ggml_tensor * tensor);
GGML_API int ggml_bitnet_get_type_bits(enum ggml_type type);
GGML_API void ggml_bitnet_set_n_threads(int n_threads);
#if defined(GGML_BITNET_ARM_TL1)
GGML_API void ggml_qgemm_lut(int m, int k, void* A, void* LUT, void* Scales, void* LUT_Scales, void* C);
GGML_API void ggml_preprocessor(int m, int k, void* B, void* LUT_Scales, void* QLUT);
#endif
#if defined(GGML_BITNET_X86_TL2)
GGML_API void ggml_qgemm_lut(int bs, int m, int k, int BK, void* A, void* sign, void* LUT, void* Scales, void* LUT_Scales, void* C);
GGML_API void ggml_preprocessor(int bs, int m, int three_k, int two_k, void* B, void* LUT_Scales, void* Three_QLUT, void* Two_QLUT);
#endif

Start by tracing call sites for the listed ggml_bitnet_* declarations and the llama.cpp inference flow; the payload does not name specific files or entry points. Document where initialization, capability checks, tensor transformation, and matrix-multiplication functions enter the flow, including the conditional ARM and x86 paths. Done means the integration path is clearly explained.