From help-request at octave dot org Fri Oct 8 12:54:35 2004 Subject: Re: octave-2.1.59 and Cygwin From: "Paul Thomas" To: "Yadin Y. Goldschmidt" Cc: Date: Fri, 8 Oct 2004 19:53:30 +0200 This is a multi-part message in MIME format. ------=_NextPart_000_0057_01C4AD70.7C6765F0 Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: quoted-printable Yadin, The slowness remains in all versions after gcc-3.2.x with Cygwin, as far = as I know. It is entered as a bug with the gcc developers, if you care to track it = down ( I cannot lay hands on the Bugzilla number at the moment, nor the = correspondence that I have had on it.). The cause appears to be = somewhere within the depths of new/delete, with sjlj being the likely = culprit. Implementing dwarf2 is said to be the cure in the long term. = The developers seemed to be distressed that there was a 20% difference = in the Linux version and ignored the factors of 10 or so in the Cygwin = variety. It is easy to demonstrate that new/delete is slow in the later versions = of gcc. To be perfectly honest, I did not understand the explanations = that were offered and I am left perturbed that old versions of gcc are = just fine. I had wondered whether the old version of new/delete could = be bolted on to the later gcc but I simply have not had the time to = explore this. In the meantime, perhaps we should post the gcc-3.2 Cygwin binary = somewhere? A number of us have copies. Paul T ------=_NextPart_000_0057_01C4AD70.7C6765F0 Content-Type: text/html; charset="iso-8859-1" Content-Transfer-Encoding: quoted-printable
Yadin,
 
The slowness remains in all versions = after=20 gcc-3.2.x with Cygwin, as far as I know.
 
It is entered as a bug with the gcc = developers, if=20 you care to track it down ( I cannot lay hands on the Bugzilla number at = the=20 moment, nor the correspondence that I have had on it.).  The cause = appears=20 to be somewhere within the depths of new/delete, with sjlj being the = likely=20 culprit. Implementing dwarf2 is said to be the cure in the long = term. =20 The developers seemed to be distressed that there was a 20% difference = in the=20 Linux version and ignored the factors of 10 or so in the Cygwin=20 variety.
 
It is easy to demonstrate that = new/delete is slow=20 in the later versions of gcc.  To be perfectly honest, I did not = understand=20 the explanations that were offered and I am left perturbed that old = versions of=20 gcc are just fine.  I had wondered whether the old version of = new/delete=20 could be bolted on to the later gcc but I simply have not had the time = to=20 explore this.
 
In the meantime, perhaps we should post = the gcc-3.2=20 Cygwin binary somewhere?  A number of us have copies.
 
Paul T
------=_NextPart_000_0057_01C4AD70.7C6765F0-- ------------------------------------------------------------- Octave is freely available under the terms of the GNU GPL. Octave's home on the web: http://www.octave.org How to fund new projects: http://www.octave.org/funding.html Subscription information: http://www.octave.org/archive.html -------------------------------------------------------------