On Wed, Sep 23, 2009 at 11:30 AM, Evan Martin <e...@chromium.org> wrote:
>
> On Wed, Sep 23, 2009 at 11:26 AM, Mark Mentovai <m...@chromium.org> wrote:
> > If a test exposes a case where something requires 53/64-bit IEEE
> > double precision as opposed to 64/80-bit double extended, and we're
> > changing our configuration "to make tests pass" but then releasing in
> > another configuration where those same tests wouldn't pass, we're
> > doing something wrong. Using SSE2 floating-point operations in a
> > configuration that we test and then using x87 floating-point
> > operations in a configuration that we release is completely bogus.
>
> I agree; hence the bug.
>
To be fair, the bug has been around for 6 months. So it's not like it's a
"temporary" hack.
On Wed, Sep 23, 2009 at 11:33 AM, Adam Langley <a...@chromium.org> wrote:
>
> On Wed, Sep 23, 2009 at 11:26 AM, Mark Mentovai <m...@chromium.org> wrote:
> > doing something wrong. Using SSE2 floating-point operations in a
> > configuration that we test and then using x87 floating-point
> > operations in a configuration that we release is completely bogus.
>
> The reality of the situation:
>
> * x87 doubles are 80-bits in registers and 64-bits in memory. This
> means that changing the optimisation flags of the compiler (and thus
> the spill profile of the code) changes rounding errors.
> * So, using x87 means that we would need two sets of image baselines:
> debug and release. It also means that different versions of GCC might
> end up needing their own baselines.
> * However, some people don't have SSE2 processors, so requiring it in
> Chrome branded builds would exclude them.
>
> I believe the current situation, impure as it is, is the best answer.
>
I agree that this one case seems fine, but are we positive there are no
other differences in how the two apps behave when compiled with this flag?
--~--~---------~--~----~------------~-------~--~----~
Chromium Developers mailing list: chromium-dev@googlegroups.com
View archives, change email options, or unsubscribe:
http://groups.google.com/group/chromium-dev
-~----------~----~----~----~------~----~------~--~---
