chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 11 Packages Projects Releases Wiki Activity

[FIX] Correct JIT kernel compilation on newer GPUs with outdated driver metadata. (#18496)

Browse Source 
Co-authored-by: Xiaoyu Zhang <35585791+BBuf@users.noreply.github.com>
This commit is contained in:
muse-coder
2026-02-15 12:14:39 +08:00
committed by GitHub
parent 1ce3420784
commit 91230dcca8
2 changed files with 31 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
python/sglang/jit_kernel/include/sgl_kernel/utils.cuh
View File

@@ -46,7 +46,7 @@ inline constexpr auto kFullMask = 0xffffffffu;
template <bool kUsePDL>
SGL_DEVICE void PDLWaitPrimary() {
#ifndef USE_ROCM
#if !defined(USE_ROCM) && defined(__CUDA_ARCH__) && (__CUDA_ARCH__ >= 900)
if constexpr (kUsePDL) {
asm volatile("griddepcontrol.wait;" ::: "memory");
}
@@ -55,7 +55,7 @@ SGL_DEVICE void PDLWaitPrimary() {
template <bool kUsePDL>
SGL_DEVICE void PDLTriggerSecondary() {
#ifndef USE_ROCM
#if !defined(USE_ROCM) && defined(__CUDA_ARCH__) && (__CUDA_ARCH__ >= 900)
if constexpr (kUsePDL) {
asm volatile("griddepcontrol.launch_dependents;" :::);
}