format.h

// Copyright (c) OpenMMLab. All rights reserved.

#pragma once

#include <cstdint>
#include <cuda_fp16.h>
#include <cuda_runtime.h>

namespace turbomind {

void reformat_s4_k8_m(uint32_t* dst, const uint32_t* src, int m, int k, cudaStream_t st = {});

void reformat_s4_k_m8(uint32_t* dst, const uint32_t* src, int m, int k, cudaStream_t st = {});

void convert_s4_k_m8(uint32_t*       A_dst,
                     half2*          Q_dst,
                     half*           workspace,
                     const uint32_t* A_src,
                     const half*     scales,
                     const uint32_t* qzeros,
                     int             m,
                     int             k,
                     int             group_size,
                     cudaStream_t    st = {});

void convert_s4_k_m8_(uint32_t*       A_dst,
                     half2*          Q_dst,
                     half*           workspace,
                     const uint32_t* A_src,
                     const half*     scales,
                     const uint32_t* qzeros,
                     int             m,
                     int             k,
                     int             group_size,
                     cudaStream_t    st = {});                     

void transpose_qk_s4_k_m8_hf(uint32_t* dst, const uint32_t* src, int m, int k, int size_per_head, cudaStream_t st = {});

void fuse_w1_w3_s4_k_m8(uint32_t* dst, const uint32_t* src, int m, int k, cudaStream_t st = {});

void dequantize_s4(uint4* dst, const uint32_t* src, size_t count, cudaStream_t st = {});

}  // namespace turbomind