From 8783a25debb44ffb1fa0df722be2924b8df2a119 Mon Sep 17 00:00:00 2001
From: biondizzle <biondizzle@gmail.com>
Date: Thu, 28 May 2026 05:11:11 +0000
Subject: [PATCH] fix: guard cuda_bf16.h with __CUDA_ARCH__

---
 dsv4/kernels/attention/fmha_sm100.cuh | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/dsv4/kernels/attention/fmha_sm100.cuh b/dsv4/kernels/attention/fmha_sm100.cuh
index 8a01b707..f64cad8e 100644
--- a/dsv4/kernels/attention/fmha_sm100.cuh
+++ b/dsv4/kernels/attention/fmha_sm100.cuh
@@ -30,9 +30,15 @@
 #pragma once
 
 #include <cuda_runtime.h>
-#include <cuda_bf16.h>
 #include <cstdint>
 
+// __nv_bfloat16 is a built-in type on CUDA 13+
+// cuda_bf16.h may have C++17 compatibility issues with some CUDA versions;
+// only include it in device code where it's guaranteed to work.
+#if defined(__CUDA_ARCH__)
+#include <cuda_bf16.h>
+#endif
+
 // CUTLASS C++ includes (CUDA device code only)
 #if defined(__CUDA_ARCH__)
 #include <cutlass/cutlass.h>