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

[JIT Kernel]Support fused_add_rmsnorm in JIT Kernel (#17677)

Browse Source
This commit is contained in:
Qi Yuhang
2026-01-29 09:29:59 +08:00
committed by GitHub
parent 09a9147f59
commit 0368ddf9ea
5 changed files with 336 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
python/sglang/jit_kernel/include/sgl_kernel/vec.cuh
View File

@@ -45,7 +45,7 @@ template <typename T, std::size_t N>
struct AlignedVector {
private:
/// NOTE: 1. must be pow of two 2. 16 * 8 = 128 byte, which is the max vector size supported by most devices
static_assert((N > 0 && (N & (N - 1)) == 0) && sizeof(T) * N <= 16, "CUDA only support at most 128B vector op");
static_assert((N > 0 && (N & (N - 1)) == 0) && sizeof(T) * N <= 32, "CUDA only support at most 256B vector op");
using element_t = typename details::sized_int<T>;
using storage_t = AlignedStorage<element_t, N>;