* [PATCH] Regenerate ULPs for i486 builds.
@ 2016-10-28 3:02 Carlos O'Donell
2016-10-28 12:48 ` Joseph Myers
0 siblings, 1 reply; 2+ messages in thread
From: Carlos O'Donell @ 2016-10-28 3:02 UTC (permalink / raw)
To: Joseph S. Myers, GNU C Library
Joseph,
On i486 builds I see 3 failures:
FAIL: math/test-ildouble
FAIL: math/test-ldouble
FAIL: math/test-ldouble-finite
Summary of test results:
3 FAIL
2553 PASS
45 XFAIL
They are all ULPs related e.g.
~~~
Failure: Test: pow_towardzero (0x1.000002p+0, -0x2p+0)
Result:
is: 9.99999761581463530839e-01 0xf.ffffc00000bfffd0p-4
should be: 9.99999761581463530947e-01 0xf.ffffc00000bffff0p-4
difference: 1.08420217248550443400e-19 0x8.0000000000000000p-66
ulp : 2.0000
max.ulp : 1.0000
~~~
Why might we be 1-3 ULP worse than the previously checked in baseline?
The baseline were accurate when you added them in 2012, but perhaps
compiler changes have caused this regerssion.
It's not entirely out of line with what we have for the other pow
baselines, so I'm going to suggest updating the baseline, to avoid the
failures.
Regenerated with `make regen-ulps`.
OK to checkin?
2016-10-27 Carlos O'Donell <carlos@redhat.com>
* sysdeps/i386/fpu/libm-test-ulps: Regenerate.
diff --git a/sysdeps/i386/fpu/libm-test-ulps b/sysdeps/i386/fpu/libm-test-ulps
index 84da475..1858c06 100644
--- a/sysdeps/i386/fpu/libm-test-ulps
+++ b/sysdeps/i386/fpu/libm-test-ulps
@@ -1901,14 +1901,14 @@ ldouble: 4
Function: "pow_towardzero":
double: 1
idouble: 1
-ildouble: 1
-ldouble: 1
+ildouble: 4
+ldouble: 4
Function: "pow_upward":
double: 1
idouble: 1
-ildouble: 2
-ldouble: 2
+ildouble: 4
+ldouble: 4
Function: "sin":
float: 1
---
--
Cheers,
Carlos.
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: [PATCH] Regenerate ULPs for i486 builds.
2016-10-28 3:02 [PATCH] Regenerate ULPs for i486 builds Carlos O'Donell
@ 2016-10-28 12:48 ` Joseph Myers
0 siblings, 0 replies; 2+ messages in thread
From: Joseph Myers @ 2016-10-28 12:48 UTC (permalink / raw)
To: Carlos O'Donell; +Cc: GNU C Library
On Thu, 27 Oct 2016, Carlos O'Donell wrote:
> Why might we be 1-3 ULP worse than the previously checked in baseline?
Because, for the most part, when new tests or libm changes require ulps
updates, people test for i686/fpu/multiarch and so update that file but
not this one (the two can have slightly different results for some
functions because of the different implementations used). The new tests
requiring this update were from commit
c898991d8bcfacc825097ba389ffccc5367c2b2d.
> OK to checkin?
OK.
--
Joseph S. Myers
joseph@codesourcery.com
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2016-10-28 12:48 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2016-10-28 3:02 [PATCH] Regenerate ULPs for i486 builds Carlos O'Donell
2016-10-28 12:48 ` Joseph Myers
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).