Re: Stupid Intel question
Re: Stupid Intel question
- Subject: Re: Stupid Intel question
- From: Steve Checkoway <email@hidden>
- Date: Fri, 17 Mar 2006 14:02:44 -0800
On Mar 17, 2006, at 12:44 PM, Urs Heckmann wrote:
Hiya,
The port of AUs to x86 was quite easy, now I'm into optimization.
After having had some decent trouble to get -fstrict-aliasing and -
ftree-vectorize do anything useful (both of them confused type-
casted pointers in some of my optimization strategies, arrrgh,
which doesn't happen on PowerPC nor with Intel and Microsoft
compilers for x86), I've started profiling with Sharc.
As was just discussed on the Xcode-users list, -ftree-vectorize
enables -fstrict-aliasing which is breaking code in the CoreAudio
SDK. To quote Eric Albert:
We turn off strict aliasing by default because, as you've noticed,
it breaks a lot of code (even though it's more correct and results
in faster code). But we explicitly turn it on for -O2 and higher
when -ftree-vectorize is passed to the compiler, as you're doing
here. I'm not sure why, but it doesn't seem right to me,
particularly when -fno-strict-aliasing is passed.
It's interesting to note that you're having problems on Intel but not
PowerPC as it was the other way around in the thread "GCC 4 and
strict aliasing" in the Xcode list.
I know this wasn't much help, but it seems that for now you should
probably not use -ftree-vectorize nor -fstrict-aliasing, at least
until the code and/or the compiler gets fixed.
- Steve
Attachment:
smime.p7s
Description: S/MIME cryptographic signature
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Coreaudio-api mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
This email sent to email@hidden