diff --git a/src/regression/fpclassify-invalid-ld80.c b/src/regression/fpclassify-invalid-ld80.c
new file mode 100644
index 0000000..b97fe9c
--- /dev/null
+++ b/src/regression/fpclassify-invalid-ld80.c
@@ -0,0 +1,63 @@
+// commit: f657fe4b9f734d7fdea515af8dffbf7c28ce4fbc 2013-09-05
+// classify invalid x86 ld80 representations (this is ub, we follow the fpu)
+#include <math.h>
+#include <float.h>
+#include <stdint.h>
+#include "test.h"
+
+#if LDBL_MANT_DIG==64
+static char *strclass(int c)
+{
+#define C(n) case n: return #n;
+	switch (c) {
+	C(FP_NAN)
+	C(FP_INFINITE)
+	C(FP_ZERO)
+	C(FP_SUBNORMAL)
+	C(FP_NORMAL)
+	}
+	return "invalid";
+}
+
+#define T(desc,got,want) do{ \
+if (got!=want) t_error("fpclassify(%s) failed: got %s want %s\n", desc, strclass(got), #want); \
+}while(0)
+
+int main(void)
+{
+	union {
+		long double f;
+		struct {
+			uint64_t m;
+			uint16_t se;
+		} i;
+	} u;
+	int c;
+
+	u.f = 0;
+	u.i.m = (uint64_t)1<<63;
+	c = fpclassify(u.f);
+	T("zero with msb set", c, FP_NORMAL);
+	u.i.m++;
+	c = fpclassify(u.f);
+	T("subnormal with msb set", c, FP_NORMAL);
+	u.f=1;
+	u.i.m=0;
+	c = fpclassify(u.f);
+	T("normal with msb unset", c, FP_NAN);
+	u.f=INFINITY;
+	u.i.m=0;
+	c = fpclassify(u.f);
+	T("infinity with msb unset", c, FP_NAN);
+	u.f=NAN;
+	u.i.m&=(uint64_t)-1/2;
+	c = fpclassify(u.f);
+	T("nan with msb unset", c, FP_NAN);
+	return t_status;
+}
+#else
+int main(void)
+{
+	return 0;
+}
+#endif