From 3bee73207e616b1ce83db35e5f42fb6add1f427a Mon Sep 17 00:00:00 2001
From: Claude <claude@anthropic.com>
Date: Fri, 14 Aug 2026 16:33:29 +0000
Subject: [PATCH] fix: add cuda_runtime.h to hgemm_bind.cpp for cudaStream_t

---
 ex_engine/xllm_kernels/cuda/bindings/hgemm_bind.cpp | 1 +
 1 file changed, 1 insertion(+)

diff --git a/ex_engine/xllm_kernels/cuda/bindings/hgemm_bind.cpp b/ex_engine/xllm_kernels/cuda/bindings/hgemm_bind.cpp
index af94c952..f1a6028e 100644
--- a/ex_engine/xllm_kernels/cuda/bindings/hgemm_bind.cpp
+++ b/ex_engine/xllm_kernels/cuda/bindings/hgemm_bind.cpp
@@ -5,6 +5,7 @@
 //   moe_expert_gemm(input, weights, expert_counts) → output
 
 #include <torch/extension.h>
+#include <cuda_runtime.h>
 #include <cuda_fp16.h>
 #include <vector>