amd64/112222: 32-bit libc incorrectly converts some FP numbers

[Peter Jeremy]

>Number:         112222
>Category:       amd64
>Synopsis:       32-bit libc incorrectly converts some FP numbers
>Confidential:   no
>Severity:       critical
>Priority:       high
>Responsible:    freebsd-amd64
>State:          open
>Quarter:        
>Keywords:       
>Date-Required:
>Class:          sw-bug
>Submitter-Id:   current-users
>Arrival-Date:   Sun Apr 29 00:00:12 GMT 2007
>Closed-Date:
>Last-Modified:
>Originator:     Peter Jeremy
>Release:        FreeBSD 6.2-STABLE amd64
>Organization:
n/a
>Environment:
System: FreeBSD turion.vk2pj.dyndns.org 6.2-STABLE FreeBSD 6.2-STABLE #25: Tue Jan 30 05:01:57 EST 2007 root at turion.vk2pj.dyndns.org:/usr/obj/usr/src/sys/turion amd64

>Description:
	During testing of some 32-bit code on amd64, nonsense was printed
	for some floating point conversions, eg ':e+22' instead of 1e+23
	or ':e-07' instead of 1e-06.

	Further testing has localised the problem to dtoa(), though it is
	not yet clear whether the problem is in the source code, the
	compiler or the 32-bit cross-build part of buildworld.

	The problem only appears to affect numbers with a decimal fractional
	part of 1 and decimal exponents of -307, -298, -283, -280, -278, -277,
	-275, -274, -272, -271, -269, -268, -260, -258, -145, -96, -94, -93,
	-88, -85, -82, -81, -80, -77, -76, -75, -74, -72, -71, -69, -67, -66,
	-65, -64, -49, -40, -39, -36, -34, -29, -28, -24, -23, -20, -19, -16,
	-14, -12, -11, -7, -6, +23, +24, +28, +29, +31, +35, +36, +38, +40,
	+41, +42, +43, +46, +47, +48, +51, +52, +53, +55, +57, +59, +62, +65,
	+66, +67, +68, +72, +73, +74, +77, +78, +79, +80, +82, +85, +86, +87,
	+88, +89, +90, +94, +95, +96, +97, +98, +100, +103, +104, +108, +110,
	+113, +114, +115, +116, +117, +121, +122, +129, +130, +132, +133, +136,
	+137, +138, +139, +140, +141, +142, +143, +144, +145, +148, +149, +151,
	+152, +153, +154, +155, +172, +174, +178, +182, +190, +191, +192, +193,
	+198, +201, +205, +206, +207, +209, +216, +218, +221, +222, +223, +227,
	+230, +243, +244, +246, +252, +262, +263, +267, +268, +271, +273, +275,
	+277, +278, +283, +285, +288, +289, +290, +292, +294, +297, +299, +300,
	+301, +302, +303, +306.

	Linking against libc/gdtoa_dtoa.o compiled in an i386 environment
	does not show the problem.

>How-To-Repeat:
	echo 'main(){printf("%g %g %g\\n", 1e-24, 1e-7, 1e100);}' > x.c
	cc -m32 x.c
	./a.out

>Fix:
	Unknown.
>Release-Note:
>Audit-Trail:
>Unformatted: