> > So we either have the make FP_ILOGB0 machine dependent, or change some > of the implementations.
Here is my patch for the first option, tested on arm64. John Index: include/math.h =================================================================== RCS file: /cvs/src/include/math.h,v retrieving revision 1.36 diff -u -r1.36 math.h --- include/math.h 10 Mar 2018 20:52:58 -0000 1.36 +++ include/math.h 30 Oct 2020 21:31:55 -0000 @@ -56,8 +56,14 @@ #define FP_NORMAL 0x04 #define FP_SUBNORMAL 0x08 #define FP_ZERO 0x10 - -#define FP_ILOGB0 (-INT_MAX) +/* + * ISO C permits either value; choose based on arch + */ +#if defined(__i386) || defined(__amd64) +#define FP_ILOGB0 (INT_MIN) +#else +#define FP_ILOGB0 (-INT_MAX) +#endif #define FP_ILOGBNAN INT_MAX #ifdef __FP_FAST_FMA Index: lib/libm/man/ilogb.3 =================================================================== RCS file: /cvs/src/lib/libm/man/ilogb.3,v retrieving revision 1.7 diff -u -r1.7 ilogb.3 --- lib/libm/man/ilogb.3 26 Apr 2016 19:49:22 -0000 1.7 +++ lib/libm/man/ilogb.3 30 Oct 2020 21:31:55 -0000 @@ -56,7 +56,7 @@ and .Fn ilogb 0 returns -.Dv INT_MIN . +.Dv FP_ILOGB0 . The .Fn ilogbf function is a single precision version of
