diff --git a/setup.py b/setup.py
index 8ececfc..13f012f 100644
--- a/setup.py
+++ b/setup.py
@@ -13,6 +13,7 @@ cxx_flags = ['-std=c++17', '-O3', '-fPIC', '-Wno-psabi', '-Wno-deprecated-declar
 sources = ['csrc/python_api.cpp']
 build_include_dirs = [
     f'{CUDA_HOME}/include',
+    f'{CUDA_HOME}/include/cccl',
     'deep_gemm/include',
     'third-party/cutlass/include',
     'third-party/fmt/include',
diff --git a/third-party/cutlass b/third-party/cutlass
index b244379..a49a78f 160000
--- a/third-party/cutlass
+++ b/third-party/cutlass
@@ -1 +1 @@
-Subproject commit b244379d9b15574e07b73b814b88bd2233f0b3ce
+Subproject commit a49a78ffefc86a87160dfe0ccc3a3a2d1622c918