From 81d3c3ce77120b30e62eb7960aa5c65cf972d8ec Mon Sep 17 00:00:00 2001
From: Raiden Makoto <81530826+Raiden-Makoto@users.noreply.github.com>
Date: Fri, 26 Jun 2026 23:29:19 -0700
Subject: [PATCH] [AMD] [GLM5] Guard cuda_runtime.h for ROCm in
 fused_metadata_copy (#29373)

Co-authored-by: Raiden-Makoto <Raiden-Makoto@users.noreply.github.com>
---
 .../jit_kernel/csrc/elementwise/fused_metadata_copy.cuh       | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/python/sglang/jit_kernel/csrc/elementwise/fused_metadata_copy.cuh b/python/sglang/jit_kernel/csrc/elementwise/fused_metadata_copy.cuh
index 7e1f91dcf..1758c38da 100644
--- a/python/sglang/jit_kernel/csrc/elementwise/fused_metadata_copy.cuh
+++ b/python/sglang/jit_kernel/csrc/elementwise/fused_metadata_copy.cuh
@@ -35,7 +35,11 @@
 #include <tvm/ffi/container/tensor.h>
 
 #include <algorithm>  // for std::min
+#ifndef USE_ROCM
 #include <cuda_runtime.h>
+#else
+#include <hip/hip_runtime.h>
+#endif
 
 // Forward mode enum (must match Python ForwardMode in sglang/srt/layers/attention/dsa_backend.py)
 enum ForwardModeEnum { DECODE = 0, TARGET_VERIFY = 1, DRAFT_EXTEND = 2 };