fix inefficiency of math.h isless, etc. macros

previously, everything was going through an intermediate conversion to
long double, which caused the extern __fpclassifyl function to get
invoked, preventing virtually all optimizations of these operations.

with the new code, tests on constant float or double arguments compile
to a constant 0 or 1, and tests on non-constant expressions are
efficient. I may later add support for __builtin versions on compilers
that support them.

diff --git a/include/math.h b/include/math.h
index 906914011d7c09e948559488446c50847f2b03ce..b44738d75313cf073a042fd7fd5dbbbfc1642c2a 100644 (file)
--- a/include/math.h
+++ b/include/math.h
@@ -85,21 +85,36 @@ int __signbitl(long double);
 
 #define isunordered(x,y) (isnan((x)) ? ((void)(y),1) : isnan((y)))
 
-static __inline int __isrel(long double __x, long double __y, int __rel)
-{
-       if (isunordered(__x, __y)) return 0;
-       if (__rel==-2) return __x < __y;
-       if (__rel==2) return __x > __y;
-       if (__rel==-1) return __x <= __y;
-       if (__rel==1) return __x >= __y;
-       return __x != __y;
-}
-
-#define isless(x,y) __isrel((x), (y), -2)
-#define islessequal(x,y) __isrel((x), (y), -1)
-#define islessgreater(x,y) __isrel((x), (y), 0)
-#define isgreaterequal(x,y) __isrel((x), (y), 1)
-#define isgreater(x,y) __isrel((x), (y), 2)
+#define __ISREL_DEF(rel, op, type) \
+static __inline int __is##rel(type __x, type __y) \
+{ return !isunordered(__x,__y) && __x op __y; }
+
+__ISREL_DEF(lessf, <, float)
+__ISREL_DEF(less, <, double)
+__ISREL_DEF(lessl, <, long double)
+__ISREL_DEF(lessequalf, <=, float)
+__ISREL_DEF(lessequal, <=, double)
+__ISREL_DEF(lessequall, <=, long double)
+__ISREL_DEF(lessgreaterf, !=, float)
+__ISREL_DEF(lessgreater, !=, double)
+__ISREL_DEF(lessgreaterl, !=, long double)
+__ISREL_DEF(greaterf, >, float)
+__ISREL_DEF(greater, >, double)
+__ISREL_DEF(greaterl, >, long double)
+__ISREL_DEF(greaterequalf, >=, float)
+__ISREL_DEF(greaterequal, >=, double)
+__ISREL_DEF(greaterequall, >=, long double)
+
+#define __tg_pred_2(x, y, p) ( \
+       sizeof((x)+(y)) == sizeof(float) ? p##f(x, y) : \
+       sizeof((x)+(y)) == sizeof(double) ? p(x, y) : \
+       p##l(x, y) )
+
+#define isless(x, y)            __tg_pred_2(x, y, __isless)
+#define islessequal(x, y)       __tg_pred_2(x, y, __islessequal)
+#define islessgreater(x, y)     __tg_pred_2(x, y, __islessgreater)
+#define isgreater(x, y)         __tg_pred_2(x, y, __isgreater)
+#define isgreaterequal(x, y)    __tg_pred_2(x, y, __isgreaterequal)
 
 double      acos(double);
 float       acosf(float);