EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

change sgl_kernel WARP_SIZE to 64

Browse Source
This commit is contained in:
maxiao1
2025-11-03 10:17:53 +08:00
parent 8fc552638f
commit 75cd34d172
5 changed files with 5 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
sgl-kernel/csrc/gemm/qserve_w4a8_per_chn_gemm.cu
View File

@@ -25,7 +25,7 @@
#define INTRIN_M 16 #define INTRIN_M 16
#define INTRIN_N 16 #define INTRIN_N 16
#define INTRIN_K 32 #define INTRIN_K 32
#define WARP_SIZE 32 #define WARP_SIZE 64
#define SMEM_PAD_A 0 #define SMEM_PAD_A 0
#define SMEM_PAD_B 0 #define SMEM_PAD_B 0
#define PACK_SIZE 16 #define PACK_SIZE 16

2
sgl-kernel/csrc/gemm/qserve_w4a8_per_group_gemm.cu
View File

@@ -25,7 +25,7 @@
#define INTRIN_M 16 #define INTRIN_M 16
#define INTRIN_N 16 #define INTRIN_N 16
#define INTRIN_K 32 #define INTRIN_K 32
#define WARP_SIZE 32 #define WARP_SIZE 64
#define SMEM_PAD_A 0 #define SMEM_PAD_A 0
#define SMEM_PAD_B 0 #define SMEM_PAD_B 0
#define PACK_SIZE 16 #define PACK_SIZE 16

2
sgl-kernel/csrc/kvcacheio/transfer.cu
View File

@@ -5,7 +5,7 @@
#include <cstdint> #include <cstdint>
#ifndef USE_ROCM #ifndef USE_ROCM
#define WARP_SIZE 32 #define WARP_SIZE 64
#include "pytorch_extension_utils.h" #include "pytorch_extension_utils.h"
#else #else
#include "pytorch_extension_utils_rocm.h" #include "pytorch_extension_utils_rocm.h"

2
sgl-kernel/csrc/quantization/gguf/ggml-common.h
View File

@@ -3,7 +3,7 @@
// copied from https://github.com/ggerganov/llama.cpp/blob/b2899/ggml-common.h // copied from https://github.com/ggerganov/llama.cpp/blob/b2899/ggml-common.h
#define QK_K 256 #define QK_K 256
#define K_QUANTS_PER_ITERATION 2 #define K_QUANTS_PER_ITERATION 2
#define WARP_SIZE_GGUF 32 #define WARP_SIZE_GGUF 64
#define K_SCALE_SIZE 12 #define K_SCALE_SIZE 12
#define CUDA_DEQUANTIZE_BLOCK_SIZE 256 #define CUDA_DEQUANTIZE_BLOCK_SIZE 256
#define CUDA_QUANTIZE_BLOCK_SIZE 256 #define CUDA_QUANTIZE_BLOCK_SIZE 256

2
sgl-kernel/include/utils.h
View File

@@ -340,7 +340,7 @@ inline bool getEnvEnablePDL() {
#define CEILDIV(x, y) (((x) + (y) - 1) / (y)) #define CEILDIV(x, y) (((x) + (y) - 1) / (y))
#ifndef USE_ROCM #ifndef USE_ROCM
#define WARP_SIZE 32 #define WARP_SIZE 64
#else #else
#if defined(__GFX9__) || !defined(__HIP_DEVICE_COMPILE__) #if defined(__GFX9__) || !defined(__HIP_DEVICE_COMPILE__)
#define WARP_SIZE 64 #define WARP_SIZE 64