Large errors found in the GNU libc using mpcheck

Note: for each function, only the largest known errors (in ulps) are given. The error is positive when the GNU libc result is larger than the correct result, and negative otherwise.

The details the goals for functions expected to be within a few ulps are in the GNU libc manual, section Errors in Math Functions.

Single precision

See also: 14473 (cpowf)

glibc version	function	arguments	rounding mode	error in ulps	test program	bugzilla
2.18.90	erf	0x1.c5bf94p-127	RNDZ	spurious underflow	test4.c	16516
2.18.90	exp2	-0x1.f8002ep+6	RNDU	missing underflow	test8.c	16521
2.18.90	sinh	0x1.4p-146	RNDN	missing underflow	test6.c	16519
2.18.90	tan	0x1p-149	RNDZ	missing underflow	test5.c	16517
2.18.90	tanh	0x1p-149	RNDN	missing underflow	test7.c	16520
2.18.90	yn	42, 0x1.0c2d26p+9	RNDZ	-62183	test1.c

Double precision

glibc version	function	argument	rounding mode	error in ulps	test program	bugzilla
2.18.90	sincos	0x1p-1074	RNDN	missing underflow		16526
2.18.90	y0	0x1.c98266f77a21ep-1	RNDN	-22684	test0.c	16492

Long Double precision

glibc version	function	argument	rounding mode	error in ulps	test program	bugzilla
2.18.90	sin	0x1p-16445	RNDZ	missing underflow		16538
2.18.90	y0	0xe.4c120b3ec76c99dp-4	RNDU	64899	test2.c
2.18.90	yn	17, 0xd.563c0f9f9f83a9cp+6	RNDU	745252115	test3.c	16507

Note: Anything 10 ulps or more, in any rounding mode, should be reported to Bugzilla. See the meta-bug 14759.