Support compile sgl-kernel on cuda 13.0 (#9721)

This commit is contained in:
Rain Jiang
2025-08-28 10:18:03 -07:00
committed by GitHub
parent 07c9d8fba2
commit 6b39f9cf8c
13 changed files with 78 additions and 14 deletions

View File

@@ -78,7 +78,7 @@ FetchContent_Populate(repo-triton)
FetchContent_Declare(
repo-flashinfer
GIT_REPOSITORY https://github.com/flashinfer-ai/flashinfer.git
GIT_TAG 9220fb3443b5a5d274f00ca5552f798e225239b7
GIT_TAG 018b551825c8e5579206e6eb9d3229fa679202b3
GIT_SHALLOW OFF
)
FetchContent_Populate(repo-flashinfer)
@@ -174,11 +174,28 @@ if ("${CUDA_VERSION}" VERSION_GREATER_EQUAL "12.8" OR SGL_KERNEL_ENABLE_SM100A)
list(APPEND SGL_KERNEL_CUDA_FLAGS
"-gencode=arch=compute_100,code=sm_100"
"-gencode=arch=compute_100a,code=sm_100a"
"-gencode=arch=compute_101,code=sm_101"
"-gencode=arch=compute_101a,code=sm_101a"
"-gencode=arch=compute_120,code=sm_120"
"-gencode=arch=compute_120a,code=sm_120a"
)
# refer sm_121, sm_110 and sm_101 description https://github.com/pytorch/pytorch/pull/156176
if ("${CUDA_VERSION}" VERSION_GREATER_EQUAL "13.0")
list(APPEND SGL_KERNEL_CUDA_FLAGS
"-gencode=arch=compute_103,code=sm_103"
"-gencode=arch=compute_103a,code=sm_103a"
"-gencode=arch=compute_110,code=sm_110"
"-gencode=arch=compute_110a,code=sm_110a"
"-gencode=arch=compute_121,code=sm_121"
"-gencode=arch=compute_121a,code=sm_121a"
"--compress-mode=size"
)
else()
list(APPEND SGL_KERNEL_CUDA_FLAGS
"-gencode=arch=compute_101,code=sm_101"
"-gencode=arch=compute_101a,code=sm_101a"
)
endif()
else()
list(APPEND SGL_KERNEL_CUDA_FLAGS
"-use_fast_math"
@@ -261,12 +278,6 @@ set(SOURCES
"csrc/moe/cutlass_moe/w4a8/w4a8_moe_data.cu"
"csrc/moe/cutlass_moe/w4a8/w4a8_grouped_mm_c3x.cu"
"csrc/moe/marlin_moe_wna16/ops.cu"
"csrc/moe/marlin_moe_wna16/kernel_bf16_ku4.cu"
"csrc/moe/marlin_moe_wna16/kernel_bf16_ku4b8.cu"
"csrc/moe/marlin_moe_wna16/kernel_bf16_ku8b128.cu"
"csrc/moe/marlin_moe_wna16/kernel_fp16_ku4.cu"
"csrc/moe/marlin_moe_wna16/kernel_fp16_ku4b8.cu"
"csrc/moe/marlin_moe_wna16/kernel_fp16_ku8b128.cu"
"csrc/moe/moe_align_kernel.cu"
"csrc/moe/moe_fused_gate.cu"
"csrc/moe/moe_topk_softmax_kernels.cu"