Hi again,
for more details, look at 77dce024e344. It takes the original
approach of not touching operations using opaque pointers at all into
one that allows getfield_gc(p,...) with p opaque in the short_preamble
in some specific cases.
On Tue, Oct 14, 2014 at 9:55 AM, Hakan Ardo wrote:
> On Tue
On Tue, Oct 14, 2014 at 9:26 AM, Armin Rigo wrote:
>
> I'm not sure: the logic in optimizeopt/heap.py seems to be "if it's an
> opaque pointer, and if we don't know its class, then don't move it".
> According to this logic, operations on RPython instances (like
> GUARD_CLASS) can still be moved to
Hi Hakan,
On 13 October 2014 18:59, Hakan Ardo wrote:
> Are we certain opaque pointers always refer to GC-objects?
Yes.
> Right, but mark_opaque_ptr prevents this case as well by not moving
> operations using opaque pointers into the short preamble. So I guess
> we are currently fine?
I'm not
On Mon, Oct 13, 2014 at 6:09 PM, Armin Rigo wrote:
>
> I think it should work.
I'm sure your right :)
> What I have in mind for GUARD_GCTYPE would be
> to check the GC header of the object; all GC pointers have one, with
> our own GCs --- even GcArrays. So GUARD_GCTYPE would check that the
> ob
Hi Hakan,
On 13 October 2014 13:10, Hakan Ardo wrote:
> I think the problematic case is actually when mixing with arrays.
> There will already be a guard_class protecting the getfiled in the
> short preamble, but it crashes when operating on an array. Also, there
> is no guard_arraytype. So mixin
On Mon, Oct 13, 2014 at 12:24 PM, Armin Rigo wrote:
>>
>> Yes as far as I can remember. You have ofcourse the exact same issue
>> with getarrayitem_* and friends...
>
> Ah, indeed.
>
> I'm thinking about a more involved fix, prompted by
> https://bitbucket.org/pypy/pypy/issue/1886 . Would it work
Hi Hakan,
On 13 October 2014 11:52, Hakan Ardo wrote:
> On Mon, Oct 13, 2014 at 11:45 AM, Armin Rigo wrote:
>> Hi Hakan,
>>
>> On 13 October 2014 10:11, Hakan Ardo wrote:
>>> mark_opaque_ptr is used by unrolling to prevent moving getfield_gc(p1)
>>> into the short preamble is p1 if opaque. This
