Specialize __is_extended_floating_point for FP8 types (NVIDIA#3470)

Also ensure that we actually can enable FP8 due to FP16 and BF16 requirements Co-authored-by: Michael Schellenberger Costa <[email protected]>
bernhardmgruber · Jan 22, 2025 · 43e80f7 · 43e80f7
1 parent 432a060
commit 43e80f7
Show file tree

Hide file tree

Showing 3 changed files with 34 additions and 10 deletions.
diff --git a/libcudacxx/include/cuda/std/__cccl/extended_floating_point.h b/libcudacxx/include/cuda/std/__cccl/extended_floating_point.h
@@ -25,16 +25,6 @@
 #include <cuda/std/__cccl/diagnostic.h>
 #include <cuda/std/__cccl/preprocessor.h>
 
-#if !defined(_CCCL_DISABLE_NVFP8_SUPPORT)
-#  if _CCCL_HAS_INCLUDE(<cuda_fp8.h>)
-#    define _CCCL_HAS_NVFP8() 1
-#  else
-#    define _CCCL_HAS_NVFP8() 0
-#  endif // _CCCL_CUDACC_AT_LEAST(11, 8) && _CCCL_HAS_INCLUDE(<cuda_fp8.h>)
-#else
-#  define _CCCL_HAS_NVFP8() 0
-#endif // !defined(_CCCL_DISABLE_NVFP8_SUPPORT)
-
 #if !defined(_CCCL_HAS_NVFP16)
 #  if _CCCL_HAS_INCLUDE(<cuda_fp16.h>) && (_CCCL_HAS_CUDA_COMPILER || defined(LIBCUDACXX_ENABLE_HOST_NVFP16)) \
                         && !defined(CCCL_DISABLE_FP16_SUPPORT)
@@ -49,4 +39,14 @@
 #  endif
 #endif // !_CCCL_HAS_NVBF16
 
+#if !defined(_CCCL_DISABLE_NVFP8_SUPPORT)
+#  if _CCCL_HAS_INCLUDE(<cuda_fp8.h>) && defined(_CCCL_HAS_NVFP16) && defined(_CCCL_HAS_NVBF16)
+#    define _CCCL_HAS_NVFP8() 1
+#  else
+#    define _CCCL_HAS_NVFP8() 0
+#  endif // _CCCL_HAS_INCLUDE(<cuda_fp8.h>)
+#else
+#  define _CCCL_HAS_NVFP8() 0
+#endif // !defined(_CCCL_DISABLE_NVFP8_SUPPORT)
+
 #endif // __CCCL_EXTENDED_FLOATING_POINT_H
diff --git a/libcudacxx/include/cuda/std/__type_traits/is_extended_floating_point.h b/libcudacxx/include/cuda/std/__type_traits/is_extended_floating_point.h
@@ -33,6 +33,10 @@ _CCCL_DIAG_SUPPRESS_CLANG("-Wunused-function")
 _CCCL_DIAG_POP
 #endif // _LIBCUDACXX_HAS_NVBF16
 
+#if _CCCL_HAS_NVFP8()
+#  include <cuda_fp8.h>
+#endif // _CCCL_HAS_NVFP8()
+
 _LIBCUDACXX_BEGIN_NAMESPACE_STD
 
 template <class _Tp>
@@ -71,6 +75,22 @@ _CCCL_INLINE_VAR constexpr bool __is_extended_floating_point_v<__nv_bfloat16> =
 #  endif // !_CCCL_NO_INLINE_VARIABLES
 #endif // _LIBCUDACXX_HAS_NVBF16
 
+#if _CCCL_HAS_NVFP8()
+template <>
+struct __is_extended_floating_point<__nv_fp8_e4m3> : true_type
+{};
+template <>
+struct __is_extended_floating_point<__nv_fp8_e5m2> : true_type
+{};
+
+#  ifndef _CCCL_NO_INLINE_VARIABLES
+template <>
+_CCCL_INLINE_VAR constexpr bool __is_extended_floating_point_v<__nv_fp8_e4m3> = true;
+template <>
+_CCCL_INLINE_VAR constexpr bool __is_extended_floating_point_v<__nv_fp8_e5m2> = true;
+#  endif // !_CCCL_NO_INLINE_VARIABLES
+#endif // _CCCL_HAS_NVFP8()
+
 _LIBCUDACXX_END_NAMESPACE_STD
 
 #endif // _LIBCUDACXX___TYPE_TRAITS_IS_EXTENDED_FLOATING_POINT_H
diff --git a/...est/libcudacxx/libcxx/utilities/meta/meta.unary/meta.unary.cat/is_floating_point.pass.cpp b/...est/libcudacxx/libcxx/utilities/meta/meta.unary/meta.unary.cat/is_floating_point.pass.cpp
@@ -86,6 +86,10 @@ int main(int, char**)
 #ifdef _LIBCUDACXX_HAS_NVBF16
   test_is_floating_point<__nv_bfloat16>();
 #endif // _LIBCUDACXX_HAS_NVBF16
+#if _CCCL_HAS_NVFP8()
+  test_is_floating_point<__nv_fp8_e4m3>();
+  test_is_floating_point<__nv_fp8_e5m2>();
+#endif // ())
 
   test_is_not_floating_point<short>();
   test_is_not_floating_point<unsigned short>();