--- Comment #26 from rob1weld at aol dot com 2010-04-12 01:54 ---
(In reply to comment #25)
I understand that this is INVALID because all the points raised by comment
#21.
If crlibm is better than what we have, but we cannot use it, it is the same as
if it didn't exist.
It is
--- Comment #25 from manu at gcc dot gnu dot org 2010-02-20 23:43 ---
I understand that this is INVALID because all the points raised by comment #21.
If crlibm is better than what we have, but we cannot use it, it is the same as
if it didn't exist.
--
manu at gcc dot gnu dot org
--- Comment #24 from rob1weld at aol dot com 2007-06-17 20:52 ---
http://gcc.gnu.org/gcc-4.3/changes.html#mpfropts
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32180
--- Comment #23 from rob1weld at aol dot com 2007-06-16 18:12 ---
Comment #17 From [EMAIL PROTECTED] 2007-06-13 11:30 [reply] ---
On Tue, 6 Feb 2007, Florent de Dinechin wrote:
We are the maintainers of the crlibm project, which aims at developping a
modern, correctly rounded
--- Comment #20 from rob1weld at aol dot com 2007-06-15 21:23 ---
Created an attachment (id=13709)
-- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=13709action=view)
Specific example where libm, libcrlibm, and mpfr differ
Here is a specific example of three different math libraries
--- Comment #21 from kargl at gcc dot gnu dot org 2007-06-15 22:22 ---
(In reply to comment #20)
Created an attachment (id=13709)
-- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=13709action=view) [edit]
Specific example where libm, libcrlibm, and mpfr differ
Here is a specific
--- Comment #22 from joseph at codesourcery dot com 2007-06-15 22:43
---
Subject: Re: Paranoia UCB GSL TestFloat libm tests fail
- accuracy of recent gcc math poor
On Fri, 15 Jun 2007, rob1weld at aol dot com wrote:
This is just one number. How many more could there be, how will
--- Comment #19 from rob1weld at aol dot com 2007-06-14 08:14 ---
You've shown nothing to validate that crlibm is more accurate than mpfr.
So how did you do this measurement?
Read 1st section of http://www.mpfr.org/faq.html and either
crlibm-0.18beta1.pdf or better still
--- Comment #16 from rob1weld at aol dot com 2007-06-13 07:53 ---
I did some testing on the CVS of crlibm, (it needs a few files from
crlibm-1.0beta1.tar.gz).
The new docs list these interesting features:
portable to any system implementing the ISO-C99 and IEEE-754 standards,
--- Comment #17 from joseph at codesourcery dot com 2007-06-13 11:30
---
Subject: Re: Paranoia UCB GSL TestFloat libm tests fail
- accuracy of recent gcc math poor
I previously suggested to the crlibm authors that they consider assigning
it to the FSF for contribution to
--- Comment #18 from kargl at gcc dot gnu dot org 2007-06-13 17:52 ---
The libm library is the least accurate and on average the fastest; though not
fastest for every function instance. The most accurate is always CRLibm,
sometimes it is fastest. The MPFR library is second most
--- Comment #15 from rob1weld at aol dot com 2007-06-12 17:50 ---
Correctly Rounded mathematical library
http://lipforge.ens-lyon.fr/www/crlibm/index.html
CRlibm, an efficient and proven correctly-rounded mathematical library
CRlibm is a free mathematical library (libm) which
--- Comment #12 from rob1weld at aol dot com 2007-06-07 13:42 ---
I've done some more testing.
With GNU/Linux 4.0 the file: /usr/include/bits/mathdef.h has this in it:
# if defined __FLT_EVAL_METHOD__ __FLT_EVAL_METHOD__ == 0
/* When using -mfpmath=sse, values are computed with the
--- Comment #13 from rob1weld at aol dot com 2007-06-07 13:48 ---
One other thing:
When I build and test the origonal (un-modified) Paranoia with GCC I can
compile with different flags and get different results - but these are
_similar_ flags :(
1 Defect 1 Flaw
-mmmx -msse -m3dnow
1
--- Comment #14 from rob1weld at aol dot com 2007-06-08 00:23 ---
Here are the test results. I enabled almost every possible option and all the
checking that is functional. Nearly every test passed. I diffed it with a
result from a few days ago (before the mod), I do not seem to have
--- Comment #11 from rob1weld at aol dot com 2007-06-05 17:22 ---
[EMAIL PROTECTED]
IEEE 754 does not discuss any of the functions you list above.
Comment #4 From Rob
That page is a report of the libc6 tests that are ran when the code is built
[EMAIL PROTECTED]
Compare the ...
--- Comment #7 from rob1weld at aol dot com 2007-06-04 08:58 ---
[EMAIL PROTECTED]
IEEE 754 does not discuss any of the functions you list above. In fact,
IEEE 754 places requirements on exactly one function in libm, and that is
sqrt(), which must be exact in all rounding modes.
--- Comment #8 from rob1weld at aol dot com 2007-06-04 09:07 ---
[EMAIL PROTECTED]
Can you suggest a download URL?
Mine is from ATT. I didn't save the URL. Your output is certainly a few pages
shorter than mine. There is a Paranoia test included in Ucbtest that has output
similar to
--- Comment #9 from rob1weld at aol dot com 2007-06-04 09:16 ---
There is a wiki here - your contributions are appreciated.
http://en.wikipedia.org/wiki/IEEE_754r
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32180
--- Comment #10 from kargl at gcc dot gnu dot org 2007-06-04 17:32 ---
(In reply to comment #7)
[EMAIL PROTECTED]
IEEE 754 does not discuss any of the functions you list above. In fact,
IEEE 754 places requirements on exactly one function in libm, and that is
sqrt(), which must
--- Comment #2 from rob1weld at aol dot com 2007-06-03 13:16 ---
Did GSL and Paranoia with -ffloat-store for gcc 4.3.0, same result.
Instead of the normal x87 issue it might be a libm issue since it works
with Cygwin's gcc but fails with all the Linux gcc's.
Here is something that
--- Comment #3 from rob1weld at aol dot com 2007-06-03 15:15 ---
Here is simple test for the float-store issue:
main() {
double v = 1E308;
double x = (v * v) / v;
printf(Try compiling with and without -ffloat-store\n);
printf((1E308 * 1E308) / 1E308\n);
printf(Correct output is
--- Comment #4 from rob1weld at aol dot com 2007-06-03 16:05 ---
I copied
gcc-4_3-build/i686-pc-linux-gnu/libjava/classpath/native/fdlibm/.libs/libfdlibm.a
to my current directory and instead of using -lm I used ./libfdlibm.a ...
Guess what.
I propose this simple and useful fix:
We
--- Comment #5 from kargl at gcc dot gnu dot org 2007-06-03 16:35 ---
(In reply to comment #4)
Function Alpha Generic ix86 IA64 PowerPC
acosf - -- - -
acos - -- - -
cosf 1 -1 1 1
cos
--- Comment #6 from jvdelisle at gcc dot gnu dot org 2007-06-03 21:11
---
I just ran a c version of double precision paranoia, and a single precsion f77
version with latest gcc and gfortran trunk as well as with g77 from 3.4 vintage
and in all cases I get this:
No failures, defects
--- Comment #1 from pinskia at gcc dot gnu dot org 2007-06-01 16:15 ---
Can you try with -ffloat-store, this might be the normal x87 issue (see PR
323).
--
pinskia at gcc dot gnu dot org changed:
What|Removed |Added
26 matches
Mail list logo