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

Remove annoying warnings in sgl kernel build (#9905)

Browse Source
This commit is contained in:
Lianmin Zheng
2025-09-02 20:18:25 -07:00
committed by GitHub
parent 37565b7f21
commit d631290e32
5 changed files with 43 additions and 36 deletions
Download Patch File Download Diff File Expand all files Collapse all files

66
sgl-kernel/CMakeLists.txt
View File

@@ -3,6 +3,7 @@ project(sgl-kernel LANGUAGES CXX CUDA)
# CMake
cmake_policy(SET CMP0169 OLD)
cmake_policy(SET CMP0177 NEW)
include(${CMAKE_CURRENT_LIST_DIR}/cmake/utils.cmake)
set(CMAKE_COLOR_DIAGNOSTICS ON)
set(CMAKE_VERBOSE_MAKEFILE ON CACHE BOOL "ON")
@@ -50,14 +51,7 @@ FetchContent_Declare(
)
FetchContent_Populate(repo-cutlass)
FetchContent_Declare(
repo-fmt
GIT_REPOSITORY https://github.com/fmtlib/fmt
GIT_TAG 553ec11ec06fbe0beebfbb45f9dc3c9eabd83d28
GIT_SHALLOW OFF
)
FetchContent_Populate(repo-fmt)
# DeepGEMM
FetchContent_Declare(
repo-deepgemm
GIT_REPOSITORY https://github.com/sgl-project/DeepGEMM
@@ -66,6 +60,14 @@ FetchContent_Declare(
)
FetchContent_Populate(repo-deepgemm)
FetchContent_Declare(
repo-fmt
GIT_REPOSITORY https://github.com/fmtlib/fmt
GIT_TAG 553ec11ec06fbe0beebfbb45f9dc3c9eabd83d28
GIT_SHALLOW OFF
)
FetchContent_Populate(repo-fmt)
# Triton
FetchContent_Declare(
repo-triton
@@ -148,21 +150,40 @@ set(SGL_KERNEL_CUDA_FLAGS
"--expt-extended-lambda"
"--threads=32"
# Suppress warnings
"-Xcompiler=-Wconversion"
"-Xcompiler=-fno-strict-aliasing"
# Supress warnings
"-Xcompiler=-Wno-clang-format-violations"
"-Xcompiler=-Wno-conversion"
"-Xcompiler=-Wno-deprecated-declarations"
"-Xcompiler=-Wno-terminate"
"-Xcompiler=-Wfatal-errors"
"-Xcompiler=-ftemplate-backtrace-limit=1"
"-Xcudafe=--diag_suppress=177" # variable was declared but never referenced
# uncomment to debug
# "--ptxas-options=-v"
# "--ptxas-options=--verbose,--register-usage-level=10,--warn-on-local-memory-usage"
)
option(SGL_KERNEL_ENABLE_SM100A "Enable SM100A" OFF)
option(SGL_KERNEL_ENABLE_SM90A "Enable SM90A" OFF)
option(SGL_KERNEL_ENABLE_BF16 "Enable BF16" ON)
option(SGL_KERNEL_ENABLE_FP8 "Enable FP8" ON)
option(SGL_KERNEL_ENABLE_FP4 "Enable FP4" OFF)
option(SGL_KERNEL_ENABLE_FA3 "Enable FA3" OFF)
option(SGL_KERNEL_ENABLE_SM90A "Enable SM90A" OFF)
option(SGL_KERNEL_ENABLE_SM100A "Enable SM100A" OFF)
if (SGL_KERNEL_ENABLE_BF16)
list(APPEND SGL_KERNEL_CUDA_FLAGS
"-DFLASHINFER_ENABLE_BF16"
)
endif()
if (SGL_KERNEL_ENABLE_FP8)
list(APPEND SGL_KERNEL_CUDA_FLAGS
"-DFLASHINFER_ENABLE_FP8"
"-DFLASHINFER_ENABLE_FP8_E4M3"
"-DFLASHINFER_ENABLE_FP8_E5M2"
)
endif()
if (ENABLE_BELOW_SM90)
list(APPEND SGL_KERNEL_CUDA_FLAGS
@@ -210,31 +231,12 @@ if ("${CUDA_VERSION}" VERSION_GREATER_EQUAL "12.4" OR SGL_KERNEL_ENABLE_SM90A)
)
endif()
if (SGL_KERNEL_ENABLE_BF16)
list(APPEND SGL_KERNEL_CUDA_FLAGS
"-DFLASHINFER_ENABLE_BF16"
)
endif()
if (SGL_KERNEL_ENABLE_FP8)
list(APPEND SGL_KERNEL_CUDA_FLAGS
"-DFLASHINFER_ENABLE_FP8"
"-DFLASHINFER_ENABLE_FP8_E4M3"
"-DFLASHINFER_ENABLE_FP8_E5M2"
)
endif()
if ("${CUDA_VERSION}" VERSION_GREATER_EQUAL "12.8" OR SGL_KERNEL_ENABLE_FP4)
list(APPEND SGL_KERNEL_CUDA_FLAGS
"-DENABLE_NVFP4=1"
)
endif()
string(REPLACE "-D__CUDA_NO_HALF_OPERATORS__" "" CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS}")
string(REPLACE "-D__CUDA_NO_HALF_CONVERSIONS__" "" CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS}")
string(REPLACE "-D__CUDA_NO_BFLOAT16_CONVERSIONS__" "" CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS}")
string(REPLACE "-D__CUDA_NO_HALF2_OPERATORS__" "" CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS}")
set(SOURCES
"csrc/allreduce/custom_all_reduce.cu"
"csrc/allreduce/mscclpp_allreduce.cu"