[libc-commits] [libc] [libc] Add __FE_DENORM to the fenv macros (PR #91353)

[Matt Arsenault via libc-commits]

================
@@ -9,11 +9,12 @@
 #ifndef LLVM_LIBC_MACROS_FENV_MACROS_H
 #define LLVM_LIBC_MACROS_FENV_MACROS_H
 
-#define FE_DIVBYZERO 1
-#define FE_INEXACT 2
-#define FE_INVALID 4
-#define FE_OVERFLOW 8
-#define FE_UNDERFLOW 16
+#define FE_DIVBYZERO 0x1
+#define FE_INEXACT 0x2
+#define FE_INVALID 0x4
+#define FE_OVERFLOW 0x8
+#define FE_UNDERFLOW 0x10
+#define __FE_DENORM 0x20
----------------
arsenm wrote:

For AMDGPU the best values to use would just be to directly take the bit positions from TRAP_STS. i.e. INPUT_DENORMAL is bit 1, FE_INVALID is 0, and so on

https://github.com/llvm/llvm-project/pull/91353