From ce4ecba47751a26d0650e56e363b647c8fd02435 Mon Sep 17 00:00:00 2001
From: Yineng Zhang <me@zhyncs.com>
Date: Wed, 23 Apr 2025 14:17:42 -0700
Subject: [PATCH] fix: only compile ApplyTokenBitmaskInplace cu124+ (#5686)

---
 sgl-kernel/csrc/grammar/apply_token_bitmask_inplace_cuda.cu | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/sgl-kernel/csrc/grammar/apply_token_bitmask_inplace_cuda.cu b/sgl-kernel/csrc/grammar/apply_token_bitmask_inplace_cuda.cu
index 9a99debb6..b36846531 100644
--- a/sgl-kernel/csrc/grammar/apply_token_bitmask_inplace_cuda.cu
+++ b/sgl-kernel/csrc/grammar/apply_token_bitmask_inplace_cuda.cu
@@ -26,6 +26,8 @@
 #include <ATen/cuda/CUDAContext.h>
 // clang-format on
 
+#if defined CUDA_VERSION && CUDA_VERSION >= 12040
+
 #ifndef CUDART_INF_FP16
 #define CUDART_INF_FP16 __ushort_as_half((unsigned short)0x7C00U)
 #endif
@@ -249,3 +251,4 @@ void ApplyTokenBitmaskInplace(at::Tensor logits, at::Tensor bitmask, at::optiona
       break;
   }
 }
+#endif