minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[AMD] Fix JIT compile failure in sgl_kernel/warp.cuh (#33195)

Browse Source 
Co-authored-by: Bingxu Chen <bingxche@amd.com>
This commit is contained in:
Wang, FangYuan
2026-08-02 23:35:29 -07:00
committed by GitHub
co-authored by Bingxu Chen
parent e9366d7f79
commit 21d930aae3
1 changed files with 4 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/kernels/jit/include/sgl_kernel/warp.cuh
+4
View File
@@ -181,7 +181,11 @@ SGL_DEVICE uint32_t inclusive_sum(uint32_t lane_id, uint32_t val) {
static_assert(kWarpThreads == 32); static_assert(kWarpThreads == 32);
#pragma unroll #pragma unroll
for (uint32_t offset = 1; offset < 32; offset *= 2) { for (uint32_t offset = 1; offset < 32; offset *= 2) {
#ifndef USE_ROCM
uint32_t n = __shfl_up_sync(0xFFFFFFFF, val, offset); uint32_t n = __shfl_up_sync(0xFFFFFFFF, val, offset);
#else
uint32_t n = __shfl_up_sync(kFullMask, val, offset, kWarpThreads);
#endif
if (lane_id >= offset) val += n; if (lane_id >= offset) val += n;
} }
return val; return val;