* [PATCH] Update ppc ulps
@ 2006-09-24 10:13 Jakub Jelinek
2006-09-24 20:18 ` Steven Munroe
0 siblings, 1 reply; 3+ messages in thread
From: Jakub Jelinek @ 2006-09-24 10:13 UTC (permalink / raw)
To: Ulrich Drepper; +Cc: Glibc hackers
Hi!
I'm getting on both ppc and ppc64 2.125 ulp error on clog10 (> 2ulp
allowed), 0.75ulp on y0 1/8 and 1ulp on imag csinh (-2-3i).
2006-09-24 Jakub Jelinek <jakub@redhat.com>
* sysdeps/powerpc/fpu/libm-test-ulps: Updated.
--- libc/sysdeps/powerpc/fpu/libm-test-ulps.jj 2006-01-31 22:31:27.000000000 +0100
+++ libc/sysdeps/powerpc/fpu/libm-test-ulps 2006-09-24 11:55:42.000000000 +0200
@@ -264,8 +264,8 @@ ldouble: 1
Test "Real part of: clog10 (0.75 + 1.25 i) == 0.163679467193165171449476605077428975 + 0.447486970040493067069984724340855636 i":
float: 1
ifloat: 1
-ildouble: 2
-ldouble: 2
+ildouble: 3
+ldouble: 3
Test "Imaginary part of: clog10 (3 + inf i) == inf + pi/2*log10(e) i":
double: 1
float: 1
@@ -364,6 +364,8 @@ ldouble: 2
Test "Imaginary part of: csinh (-2 - 3 i) == 3.59056458998577995201256544779481679 - 0.530921086248519805267040090660676560 i":
double: 1
idouble: 1
+ildouble: 1
+ldouble: 1
Test "Real part of: csinh (0.75 + 1.25 i) == 0.259294854551162779153349830618433028 + 1.22863452409509552219214606515777594 i":
float: 1
ifloat: 1
@@ -739,6 +741,9 @@ idouble: 1
ifloat: 1
# y0
+Test "y0 (0.125) == -1.38968062514384052915582277745018693":
+ildouble: 1
+ldouble: 1
Test "y0 (0.75) == -0.137172769385772397522814379396581855":
ildouble: 1
ldouble: 1
@@ -800,6 +805,9 @@ ildouble: 2
ldouble: 2
# yn
+Test "yn (0, 0.125) == -1.38968062514384052915582277745018693":
+ildouble: 1
+ldouble: 1
Test "yn (0, 0.75) == -0.137172769385772397522814379396581855":
ildouble: 1
ldouble: 1
@@ -1066,8 +1074,8 @@ ldouble: 1
Function: Real part of "clog10":
float: 1
ifloat: 1
-ildouble: 2
-ldouble: 2
+ildouble: 3
+ldouble: 3
Function: Imaginary part of "clog10":
double: 1
Jakub
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: [PATCH] Update ppc ulps
2006-09-24 10:13 [PATCH] Update ppc ulps Jakub Jelinek
@ 2006-09-24 20:18 ` Steven Munroe
2006-09-24 20:49 ` Jakub Jelinek
0 siblings, 1 reply; 3+ messages in thread
From: Steven Munroe @ 2006-09-24 20:18 UTC (permalink / raw)
To: Jakub Jelinek; +Cc: Ulrich Drepper, Glibc hackers
Jakub Jelinek wrote:
>Hi!
>
>I'm getting on both ppc and ppc64 2.125 ulp error on clog10 (> 2ulp
>allowed), 0.75ulp on y0 1/8 and 1ulp on imag csinh (-2-3i).
>
>
>
Which gcc are you using? I found that gcc-4.2 or gcc-4.1.1 with the
reassociation patch backport causes upls errors in ppc that are not
there for just gcc-4.1.1.
I have not had time to tract this down to a simple test case that I can
hand to the gcc folks.
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: [PATCH] Update ppc ulps
2006-09-24 20:18 ` Steven Munroe
@ 2006-09-24 20:49 ` Jakub Jelinek
0 siblings, 0 replies; 3+ messages in thread
From: Jakub Jelinek @ 2006-09-24 20:49 UTC (permalink / raw)
To: Steven Munroe; +Cc: Ulrich Drepper, Glibc hackers
On Sun, Sep 24, 2006 at 03:11:45PM -0500, Steven Munroe wrote:
> >I'm getting on both ppc and ppc64 2.125 ulp error on clog10 (> 2ulp
> >allowed), 0.75ulp on y0 1/8 and 1ulp on imag csinh (-2-3i).
> >
> Which gcc are you using? I found that gcc-4.2 or gcc-4.1.1 with the
> reassociation patch backport causes upls errors in ppc that are not
> there for just gcc-4.1.1.
I have reassoc backport in. That's one of the reasons why I haven't
regenerated the ulps, just merged, so that both vanilla 4.1.x, 4.1.x
with reassoc etc. pass. While in the above 4 cases gcc with reassoc
generates worse code, for several ones there is an improvement too.
Jakub
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2006-09-24 20:49 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2006-09-24 10:13 [PATCH] Update ppc ulps Jakub Jelinek
2006-09-24 20:18 ` Steven Munroe
2006-09-24 20:49 ` Jakub Jelinek
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).