biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Kernel] fp4 marlin kernel (#17687)

Browse Source 
Signed-off-by: Jinzhen Lin <linjinzhen@hotmail.com>
This commit is contained in:
Jinzhen Lin
2025-05-11 10:58:49 +08:00
committed by GitHub
parent ca66a1674c
commit d74e5f37bc
21 changed files with 1216 additions and 331 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
csrc/quantization/gptq_marlin/kernel.h
View File

@@ -7,13 +7,14 @@
#include "marlin_dtypes.cuh"
#include "core/scalar_type.hpp"
#define MARLIN_KERNEL_PARAMS \
const int4 *__restrict__ A, const int4 *__restrict__ B, \
int4 *__restrict__ C, int4 *__restrict__ C_tmp, \
const int4 *__restrict__ scales_ptr, const int4 *__restrict__ zp_ptr, \
const int *__restrict__ g_idx, int num_groups, int prob_m, int prob_n, \
int prob_k, int lda, int *locks, bool use_atomic_add, \
bool use_fp32_reduce, int max_shared_mem
#define MARLIN_KERNEL_PARAMS \
const int4 *__restrict__ A, const int4 *__restrict__ B, \
int4 *__restrict__ C, int4 *__restrict__ C_tmp, \
const int4 *__restrict__ scales_ptr, \
const uint16_t *__restrict__ scale2_ptr, \
const int4 *__restrict__ zp_ptr, const int *__restrict__ g_idx, \
int num_groups, int prob_m, int prob_n, int prob_k, int lda, int *locks, \
bool use_atomic_add, bool use_fp32_reduce, int max_shared_mem
namespace MARLIN_NAMESPACE_NAME {
template <typename scalar_t, // compute dtype, half or nv_float16