fix FLT_ROUNDS to reflect the current rounding mode

Implemented as a wrapper around fegetround introducing a new function
to the ABI: __flt_rounds. (fegetround cannot be used directly from float.h)

arch/arm/bits/float.h

-#define FLT_ROUNDS 1
 #define FLT_EVAL_METHOD 0
 
 #define LDBL_TRUE_MIN 4.94065645841246544177e-324L

arch/i386/bits/float.h

-#define FLT_ROUNDS 1
 #ifdef __FLT_EVAL_METHOD__
 #define FLT_EVAL_METHOD __FLT_EVAL_METHOD__
 #else

arch/microblaze/bits/float.h

-#define FLT_ROUNDS 1
 #define FLT_EVAL_METHOD 0
 
 #define LDBL_TRUE_MIN 4.94065645841246544177e-324L

arch/mips/bits/float.h

-#define FLT_ROUNDS 1
 #define FLT_EVAL_METHOD 0
 
 #define LDBL_TRUE_MIN 4.94065645841246544177e-324L

arch/or1k/bits/float.h

-#define FLT_ROUNDS 1
 #define FLT_EVAL_METHOD 0
 
 #define LDBL_TRUE_MIN 4.94065645841246544177e-324L

arch/powerpc/bits/float.h

-#define FLT_ROUNDS 1
 #define FLT_EVAL_METHOD 0
 
 #define LDBL_TRUE_MIN 4.94065645841246544177e-324L

arch/sh/bits/float.h

-#define FLT_ROUNDS 1
 #define FLT_EVAL_METHOD 0
 
 #define LDBL_TRUE_MIN 4.94065645841246544177e-324L

arch/x32/bits/float.h

-#define FLT_ROUNDS 1
 #ifdef __FLT_EVAL_METHOD__
 #define FLT_EVAL_METHOD __FLT_EVAL_METHOD__
 #else

arch/x86_64/bits/float.h

-#define FLT_ROUNDS 1
 #ifdef __FLT_EVAL_METHOD__
 #define FLT_EVAL_METHOD __FLT_EVAL_METHOD__
 #else

include/float.h

 #ifndef _FLOAT_H
 #define _FLOAT_H
 
+int __flt_rounds(void);
+#define FLT_ROUNDS (__flt_rounds())
+
 #define FLT_RADIX 2
 
 #define FLT_TRUE_MIN 1.40129846432481707092e-45F

src/fenv/__flt_rounds.c

+#include <float.h>
+#include <fenv.h>
+
+int __flt_rounds()
+{
+	switch (fegetround()) {
+#ifdef FE_TOWARDZERO
+	case FE_TOWARDZERO: return 0;
+#endif
+	case FE_TONEAREST: return 1;
+#ifdef FE_UPWARD
+	case FE_UPWARD: return 2;
+#endif
+#ifdef FE_DOWNWARD
+	case FE_DOWNWARD: return 3;
+#endif
+	}
+	return -1;
+}