|
From: | Lassi Tuura |
Subject: | Re: [Libunwind-devel] libunwind x86-64 optimisations? |
Date: | Mon, 6 Jul 2009 18:59:03 +0200 |
Hi,
There will be some with i686 and use of SSE arguments in parameters; on x86-64 and use of AVX arguments in parameters.Since AVX hardware isn't released yet, you'll not yet see any real- lifeuses on x86-64, but it's coming.
Thanks.I take this is the dynamic alignment for vectorisation that was mentioned. I read up on this; would anyone have a concrete example of how the unwind info will look like? If it's some simple address mask I am sure I can plan to fold that in.
We move glacially slowly with our platform support so AVX isn't likely an issue right now, but could become so if we start using math libraries which configure themselves dynamically to the platform. (Our applications process a lot of numerical data, and vectorisation is actively studied now.)
I didn't find any use of DW_CFA_def_cfa_expression on x86-64 RHEL4, RHEL5 or Ubuntu 8.04 apart from syscalls, but I assume these are old enough systems the compiler wouldn't use it yet.
Lassi
[Prev in Thread] | Current Thread | [Next in Thread] |