On 2012/03/09 09:45:28, Yang wrote:
On 2012/03/09 09:30:59, kalmard wrote:
> On 2012/03/08 10:37:49, Yang wrote:
> > Thanks, I update my CL at http://codereview.chromium.org/9592047/
>
> Thank you. Just to clarify, we are happy with your change and would
prefer
that
> patch over this one.
>
> > Shouldn't the compiler be able to hide the complexity for aligned and
> unaligned
> > stores on MIPS?
>
> The MIPS ABI requires doubles and objects containing doubles to be
aligned
to
8
> byte boundaries. This way the compiler can just assume that memory is
properly
> aligned when writing to a double. In this specific case this assumption
is
> wrong, since HeapNumber is not 8-byte aligned.
HeapNumber::set_value inlines write_double_field (objects-inl.h), which
has a
specific implementation for MIPS. Wouldn't that be sufficient?
I just landed http://codereview.chromium.org/9592047/. Since we use
HeapNumber::set_value to write the double to memory, I guess this would
make any
additional MIPS-specific patch unnecessary?
http://codereview.chromium.org/9617011/
--
v8-dev mailing list
[email protected]
http://groups.google.com/group/v8-dev
