[prev in list] [next in list] [prev in thread] [next in thread] 

List:       kde-devel
Subject:    Re: Configure with icc
From:       Allan Sandfeld Jensen <kde () carewolf ! com>
Date:       2005-04-06 14:45:54
Message-ID: 200504061645.54605.kde () carewolf ! com
[Download RAW message or body]

On Wednesday 06 April 2005 15:44, Allen Winter wrote:
>
> Allan (nice name btw..although possibly misspelled ;>)
:P

> I too have an AMD system.   When you say "deliberately break" do you mean
> that the executable code is "slowed-down"?  or that the executable code is
> "wrong"?
>
It puts processor-checks in the beginning of the code. Officially to detect if 
the processors supports extensions that have been used, such as MMX and SSE. 
The old version 6 and 7, would just pick the slower path for AMD processors, 
which could be circumvented by only compiling one path in the binary (-xM vs 
-axM). The newer version 8 will just break and claim a AMD processor is 
incompatible. There are also workarounds for this. You can for instance 
compile only libraries with icc (remember still to use -x and not -ax), and 
compile the executables with gcc. For KDE this could almost be done by only 
compiling kdeinit with gcc and the rest with icc. 
Until recently I was still using a Qt library compiled with icc, together with 
KDE compiled with gcc 3.3.

I have to note though that I haven't used icc for more than half a year. You 
should try and check what happens if you compile an executable which can be 
vectorized with "icc -xiMW", and try and run it on your AMD.

I doubt however they have fixed it. The problem was that until they broke it 
for AMD processors, AMD was using the Intel compiler for official specmarks, 
and they hated that. With now even 64bits instructions in icc and a better 
implementation of them on AMD, they have even more reason to make sure AMD 
can't use their compiler unhacked.

`Allan
 
>> Visit http://mail.kde.org/mailman/listinfo/kde-devel#unsub to unsubscribe <<
[prev in list] [next in list] [prev in thread] [next in thread]