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

Fix gid calculation in per_tensor_absmax_kernel (#17126)

Browse Source
This commit is contained in:
cctry
2026-01-15 23:21:41 +08:00
committed by GitHub
parent d11e2dc6f4
commit dda35ccbd8
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/jit_kernel/csrc/gemm/per_tensor_quant_fp8.cuh
+1 -1
View File
@@ -23,7 +23,7 @@ per_tensor_absmax_kernel(const T* __restrict__ input, float* __restrict__ output
using namespace device; using namespace device;
constexpr uint32_t VEC_SIZE = 16 / sizeof(T); constexpr uint32_t VEC_SIZE = 16 / sizeof(T);
const int64_t gid = blockIdx.x * gridDim.x + threadIdx.x; const int64_t gid = blockIdx.x * blockDim.x + threadIdx.x;
float max_value = 0.0f; float max_value = 0.0f;
if (gid * VEC_SIZE + VEC_SIZE <= num_elements) { if (gid * VEC_SIZE + VEC_SIZE <= num_elements) {