[Bug 91320] Patch to resolve aliasing issues in src/glsl/list.h

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=91320

--- Comment #5 from tschw <tschwinger at isonews2.com> ---
Thanks for the pointer.

There are external projects using the glsl-compiler code and it'd be great to
see a fixed upstream. We're running it in the browser via emscripten where
optimized builds are crucial to achieve reasonable speed and download size.

I don't really care about the debate on principles, whether the rest of Mesa
should obey strict aliasing rules or not. I neither care whose patch fixes the
issue.

That being said, there are some advantages over the changes proposed on the
mailing list:

o My first patch already gets the job done. It may not be "clean" in terms of
language lawery, but it is very much "to the point", as I actually traced the
assembly to spot the one assumption that breaks it all. The patch can very
easily be approved to be non-breaking.

o My second patch is a 1:1 translation of the introductory inline comment into
code. Why not share our cleverness with the compiler so it can act less stupid?
It's API compatible, but allows for a optional and gradual rewrite of code that
uses 'head', 'tail', and 'tail_pred' directly as a cleanup step. Other than
PATCHv3 from the mailing list, it does not increase the memory footprint of
'exec_list' - in fact, it's binary compatible. I also added #pragma pack(1) so
its binary format is portable to ABIs that prefer an alignment of more than one
address size (x32 may be one of these, but I'm not sure).

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/intel-3d-bugs/attachments/20150713/65eea3b2/attachment.html>