From dda35ccbd85c35cdfc96ecd3b3d89fad272f4c83 Mon Sep 17 00:00:00 2001
From: cctry <shiyang@x.ai>
Date: Thu, 15 Jan 2026 07:21:41 -0800
Subject: [PATCH] Fix gid calculation in per_tensor_absmax_kernel (#17126)

---
 python/sglang/jit_kernel/csrc/gemm/per_tensor_quant_fp8.cuh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/python/sglang/jit_kernel/csrc/gemm/per_tensor_quant_fp8.cuh b/python/sglang/jit_kernel/csrc/gemm/per_tensor_quant_fp8.cuh
index 344648ffc..17c5de888 100644
--- a/python/sglang/jit_kernel/csrc/gemm/per_tensor_quant_fp8.cuh
+++ b/python/sglang/jit_kernel/csrc/gemm/per_tensor_quant_fp8.cuh
@@ -23,7 +23,7 @@ per_tensor_absmax_kernel(const T* __restrict__ input, float* __restrict__ output
   using namespace device;
   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;
   if (gid * VEC_SIZE + VEC_SIZE <= num_elements) {