[PD-dev] [ pure-data-Bugs-1692649 ] sqrt~ giving only -inf/inf
SourceForge.net
noreply at sourceforge.net
Sun Jun 17 16:39:52 CEST 2007
Bugs item #1692649, was opened at 2007-04-01 23:51
Message generated for change (Settings changed) made by eighthave
You can respond by visiting:
https://sourceforge.net/tracker/?func=detail&atid=478070&aid=1692649&group_id=55736
Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: pd-extended
Group: v0.39.2
>Status: Closed
Resolution: Postponed
Priority: 5
Private: No
Submitted By: Hans-Christoph Steiner (eighthave)
>Assigned to: Nobody/Anonymous (nobody)
Summary: sqrt~ giving only -inf/inf
Initial Comment:
For the G4 and Intel optimized builds of Pd-0.39.2-extended-test7 on Mac OS X, sqrt~ only returns -inf or inf. (Does this affect other platforms?) This is probably related to the auto-vectorization. This is not a problem on Pd-0.39.2-extended-test7 that was built for G3s.
Check the sqrt~-help.pd file for a test patch.
sqrt~ of the signal 9 returns -\infty, hence not 3 as expected. All tried int signals greater then zero returns -\infty. To test this, see either the attached patch (which is very close to the help-patch for [sqrt~]) or the help-patch for [sqrt~].
See also:
http://lists.puredata.info/pipermail/pd-list/2007-01/046266.html
----------------------------------------------------------------------
Comment By: stffn (stffn)
Date: 2007-06-16 06:14
Message:
Logged In: YES
user_id=1658640
Originator: NO
Same thing for RC3.
----------------------------------------------------------------------
Comment By: stffn (stffn)
Date: 2007-06-12 08:18
Message:
Logged In: YES
user_id=1658640
Originator: NO
In the autobuild from 12th of june it now works on mac-intel (macbook core
duo), but - sadly - not on powerpc (ibook G4). Both running Mac OS 10.4.9.
----------------------------------------------------------------------
Comment By: Hans-Christoph Steiner (eighthave)
Date: 2007-06-12 00:26
Message:
Logged In: YES
user_id=27104
Originator: YES
I removed these from the compiler flags and it seems to have fixed the
problem on Intel Macs:
-ftree-vectorize -ftree-vectorizer-verbose=2
Please confirm that this is fixed in both Intel and PowerPC.
This thread discusses the root causes of why the auto-vectorization causes
this problem:
http://lists.puredata.info/pipermail/pd-dev/2007-04/008689.html
Ideally the root of the problem would be fixed so that we can use
auto-vectorization.
----------------------------------------------------------------------
Comment By: stffn (stffn)
Date: 2007-05-05 09:11
Message:
Logged In: YES
user_id=1658640
Originator: NO
Still present in the nightly build of RC2 from 5th of May. It does however
work in the test build of .040.2-extended of the same date.
----------------------------------------------------------------------
Comment By: frey (freynz)
Date: 2007-04-18 08:06
Message:
Logged In: YES
user_id=1772500
Originator: NO
I have this same problem with pd-0.39.2-extended-rc2 on my Intel MacBook.
It doesn't appear in Miller's vanilla pd-0.40-2, on the same machine.
----------------------------------------------------------------------
Comment By: Hans-Christoph Steiner (eighthave)
Date: 2007-04-02 00:08
Message:
Logged In: YES
user_id=27104
Originator: YES
d_math.c, where sqrt.c isn't auto-vectorized, so this is probably caused
by a different optimization flag. Or maybe it's related to these
warnings:
d_math.c: In function 'init_rsqrt':
d_math.c:79: warning: dereferencing type-punned pointer will break
strict-aliasing rules
d_math.c: In function 'q8_rsqrt':
d_math.c:93: warning: dereferencing type-punned pointer will break
strict-aliasing rules
d_math.c: In function 'q8_sqrt':
d_math.c:101: warning: dereferencing type-punned pointer will break
strict-aliasing rules
----------------------------------------------------------------------
You can respond by visiting:
https://sourceforge.net/tracker/?func=detail&atid=478070&aid=1692649&group_id=55736
More information about the Pd-dev
mailing list