On 2013/09/17 12:39:27, ulan wrote:
On 2013/09/17 12:27:47, Bangfu wrote:
> On 2013/09/17 11:24:23, ulan wrote:
> > Indeed, there a bug in the current implementation.
> >
> > Could you please reupload your tests with a fix?
> >
> > One possible fix would be to replace:
> > sub(length, length, Operand(1), SetCC);
> > b(ne, &byte_loop_1);
> > // Copy bytes in word size chunks.
> > bind(&word_loop);
> >
> > with
> >
> > sub(length, length, Operand(1), SetCC);
> > b(ne, &align_loop_1);
> > // Copy bytes in word size chunks.
> > bind(&word_loop);
> >
> > And remove the unused aligh_loop label.
>
> Your suggestion will improve things, but the bug won't go away. If the
length
> reaches zero before src get aligned, assert might occur.
Right, then it is better to compare the length with kPointerSize and jump
to
byte_loop if it smaller at the very beginning instead of comparing the
length
with 0. This would also remove the need to check the length in the
align_loop.
Done.
By the way, do you want me to add more test cases for ARM macro assembler
to be
similar like X64 macro?
https://codereview.chromium.org/23480027/
--
--
v8-dev mailing list
[email protected]
http://groups.google.com/group/v8-dev
---
You received this message because you are subscribed to the Google Groups "v8-dev" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
For more options, visit https://groups.google.com/groups/opt_out.
