Brad, just to get up to date about GCC memory consumption and other GCC bugs re Gambit
Hi Brad,
About the GCC bug that sometimes GCC could take gigabytes of RAM (20-75GB) for certain Gambit files,
I re-read your previous article at http://webmail.iro.umontreal.ca/pipermail/gambit-list/2009-February/003159.h... .
(Btw the mentioning of an addition to http://dynamo.iro.umontreal.ca/wiki/index.php/Configure_script_options , I don't know if I see that addition there, you could doublecheck.)
You mention there that the problem GCC versions are 4.2.* and 4.3.* only, and that "-fno-move-loop-invariants" fixes it.
(I recall that you also reported the issue to the GCC development team and that in response they implemented a "max loops" check internally that fixed it on.. later GCC version, something like this.)
I just wanted to check with you now, that the above was all there has been to GCC's skyrocketing-RAM-consumption issue. So, if I just add CFLAGS="-fno-move-loop-invariants" to my ordinary "./configure --enable-single-host" etc. , no GCC/G++ version will skyrocket?
Are there any more gotchas to GCC use for various versions, except for that "-O2" produces broken code in some particular version?
Thanks!
Afficher les réponses par date
On 02/20/2016 04:49 PM, Adam wrote:
Hi Brad,
About the GCC bug that sometimes GCC could take gigabytes of RAM (20-75GB) for certain Gambit files,
...
Are there any more gotchas to GCC use for various versions, except for that "-O2" produces broken code in some particular version?
Well, I don't know, I don't follow all gcc development, it seems impossible to answer this question the way it's stated.
If you find a problem I'll look into it.
Brad
2016-02-23 4:23 GMT+07:00 Bradley Lucier lucier@math.purdue.edu:
On 02/20/2016 04:49 PM, Adam wrote:
Hi Brad,
About the GCC bug that sometimes GCC could take gigabytes of RAM (20-75GB) for certain Gambit files,
...
Are there any more gotchas to GCC use for various versions, except for that "-O2" produces broken code in some particular version?
Well, I don't know, I don't follow all gcc development, it seems impossible to answer this question the way it's stated.
If you find a problem I'll look into it.
Brad
Ok I'll narrow down my question:
To make Gambit work well (with --enable-single-host) on 4.2.* and 4.3.*, is "-fno-move-loop-invariants" a complete solution?
(The prob AFAIK was that it'd skyrocket to 20-75GB RAM consumption - and "-fno-move-loop-invariants" would bring it down to 1-2GB instead?)
On 02/22/2016 04:31 PM, Adam wrote:
2016-02-23 4:23 GMT+07:00 Bradley Lucier <lucier@math.purdue.edu mailto:lucier@math.purdue.edu>:
On 02/20/2016 04:49 PM, Adam wrote: Hi Brad, About the GCC bug that sometimes GCC could take gigabytes of RAM (20-75GB) for certain Gambit files, ... Are there any more gotchas to GCC use for various versions, except for that "-O2" produces broken code in some particular version? Well, I don't know, I don't follow all gcc development, it seems impossible to answer this question the way it's stated. If you find a problem I'll look into it. BradOk I'll narrow down my question:
To make Gambit work well (with --enable-single-host) on 4.2.* and 4.3.*, is "-fno-move-loop-invariants" a complete solution?
(The prob AFAIK was that it'd skyrocket to 20-75GB RAM consumption - and "-fno-move-loop-invariants" would bring it down to 1-2GB instead?)
Yes, I think so.
But why would one consider using 4.2.* and 4.3.* when much newer compilers are available.
Brad
-
Adam -
Bradley Lucier