math: add single precision error handling functions

These are supposed to be used in tail call positions when handling
special cases in new code. (fp exceptions may be raised "naturally"
by the common code path if special casing is more effort.)

This implements the error handling apis used in
https://github.com/ARM-software/optimized-routines
without errno setting.

src/internal/libm.h

@@ -216,4 +216,11 @@ extern int __signgam;
 hidden double __lgamma_r(double, int *);
 hidden float __lgammaf_r(float, int *);
 
+/* error handling functions */
+hidden float __math_xflowf(uint32_t, float);
+hidden float __math_uflowf(uint32_t);
+hidden float __math_oflowf(uint32_t);
+hidden float __math_divzerof(uint32_t);
+hidden float __math_invalidf(float);
+
 #endif

src/math/__math_divzerof.c

@@ -0,0 +1,6 @@
+#include "libm.h"
+
+float __math_divzerof(uint32_t sign)
+{
+	return fp_barrierf(sign ? -1.0f : 1.0f) / 0.0f;
+}

src/math/__math_invalidf.c

@@ -0,0 +1,6 @@
+#include "libm.h"
+
+float __math_invalidf(float x)
+{
+	return (x - x) / (x - x);
+}

src/math/__math_oflowf.c

@@ -0,0 +1,6 @@
+#include "libm.h"
+
+float __math_oflowf(uint32_t sign)
+{
+	return __math_xflowf(sign, 0x1p97f);
+}

src/math/__math_uflowf.c

@@ -0,0 +1,6 @@
+#include "libm.h"
+
+float __math_uflowf(uint32_t sign)
+{
+	return __math_xflowf(sign, 0x1p-95f);
+}

src/math/__math_xflowf.c

@@ -0,0 +1,6 @@
+#include "libm.h"
+
+float __math_xflowf(uint32_t sign, float y)
+{
+	return eval_as_float(fp_barrierf(sign ? -y : y) * y);
+}