Question about the INFINITY value

Assuming I add the relative suffix (like INFINITY##f) would decrementing it give the max value of the type? Reason being is I'm trying to extract that value out of GCC builtins:

Code:

---

#ifdef __HAVE_FLOAT16
typedef _Float16 paw16fd;
#ifdef INFINITY
#define PAW16FD_INF (__builtin_huge_inf16())
#else
#define PAW16FD_MAX (__builtin_huge_val16())
#endif
#endif

---

Related question, I extracted that __HAVE_FLOAT16 from the math.h library, is that actually a defined GCC gives or is that just something defined in one it's included headers? Haven't been able to find it yet nor any information on it.

Hello.

Quote:

---

Originally Posted by awsdert

Assuming I add the relative suffix (like INFINITY##f) would decrementing it give the max value of the type? Reason being is I'm trying to extract that value out of GCC builtins:

---

Maybe I do not fully understand your goal here, but why not make use of the XXXXXX_MAX family of macros in stddef.h, or if you need floats, HUGE_VAL from math.h? HUGE_VAL is ANSI C, so you definitely have it.

Quote:

---

Related question, I extracted that __HAVE_FLOAT16 from the math.h library, is that actually a defined GCC gives or is that just something defined in one it's included headers? Haven't been able to find it yet nor any information on it.

---

If by "is that actually a defined GCC gives" you mean a predefined macro given by the compiler, it doesn't appear to be. By the looks of it, it is defined in the floatn-common.h header from glibc, which can be found here. The header is included by floatn.h (found here, granted, this is an old repository), which finally is included by math.h on systems with glibc (albeit, other libcs might provide the macro).

Quote:

---

Originally Posted by interglobber

Hello.
Maybe I do not fully understand your goal here, but why not make use of the XXXXXX_MAX family of macros in stddef.h, or if you need floats, HUGE_VAL from math.h? HUGE_VAL is ANSI C, so you definitely have it.

---

Because the defines are not given for float16 etc as far as I've found thus far so the only way to get them is to calculate them somehow, I would rather not resort to using pointer hackery & integer decrement to achieve the same if simply decrementing INFINITY in the correct format would result in what I want anyways

Quote:

---

Originally Posted by interglobber

If by "is that actually a defined GCC gives" you mean a predefined macro given by the compiler, it doesn't appear to be. By the looks of it, it is defined in the floatn-common.h header from glibc, which can be found here. The header is included by floatn.h (found here, granted, this is an old repository), which finally is included by math.h on systems with glibc (albeit, other libcs might provide the macro).

---

Thanks, I found some other macros I'll have to use then, they're just not as flexible due to their FMA specific nature:

Code:

---

__FP_FAST_FMAF16
__FP_FAST_FMAF32
__FP_FAST_FMAF64
__FP_FAST_FMAF128
__FP_FAST_FMAF32X
__FP_FAST_FMAF64X
__FP_FAST_FMAF128X

---

Mentioned here: Common Predefined Macros (The C Preprocessor)