Hi Julian! Thanks for walking me through this.
On 2019-12-14T00:19:04+0000, Julian Brown <jul...@codesourcery.com> wrote:
> On Fri, 13 Dec 2019 16:25:25 +0100
> Thomas Schwinge <tho...@codesourcery.com> wrote:
>> On 2019-10-29T12:15:01+0000, Julian Brown <jul...@codesourcery.com>
>> wrote:
>> > static int
>> > -find_pointer (int pos, size_t mapnum, unsigned short *kinds)
>> > +find_group_last (int pos, size_t mapnum, unsigned short *kinds)
>> > {
>> > - if (pos + 1 >= mapnum)
>> > - return 0;
>> > + unsigned char kind0 = kinds[pos] & 0xff;
>> > + int first_pos = pos, last_pos = pos;
>> >
>> > - unsigned char kind = kinds[pos+1] & 0xff;
>> > -
>> > - if (kind == GOMP_MAP_TO_PSET)
>> > - return 3;
>> > - else if (kind == GOMP_MAP_POINTER)
>> > - return 2;
>> > + if (kind0 == GOMP_MAP_TO_PSET)
>> > + {
>> > + while (pos + 1 < mapnum && (kinds[pos + 1] & 0xff) ==
>> > GOMP_MAP_POINTER)
>> > + last_pos = ++pos;
>> > + /* We expect at least one GOMP_MAP_POINTER after a
>> > GOMP_MAP_TO_PSET. */
>> > + assert (last_pos > first_pos);
>> > + }
>> > + else
>> > + {
>> > + /* GOMP_MAP_ALWAYS_POINTER can only appear directly after some other
>> > + mapping. */
>> > + if (pos + 1 < mapnum
>> > + && (kinds[pos + 1] & 0xff) == GOMP_MAP_ALWAYS_POINTER)
>> > + return pos + 1;
>> > +
>> > + /* We can have one or several GOMP_MAP_POINTER mappings after a
>> > to/from
>> > + (etc.) mapping. */
>> > + while (pos + 1 < mapnum && (kinds[pos + 1] & 0xff) ==
>> > GOMP_MAP_POINTER)
>> > + last_pos = ++pos;
>> > + }
>> >
>> > - return 0;
>> > + return last_pos;
>> > }
Given:
program test
implicit none
integer, parameter :: n = 64
integer :: a(n)
call test_array(a)
contains
subroutine test_array(a)
implicit none
integer :: a(n)
!$acc enter data copyin(a)
!$acc exit data delete(a)
end subroutine test_array
end program test
..., we get a 'GOMP_MAP_TO' followed by a 'GOMP_MAP_POINTER'. That got
us 'find_pointer () == 2', and now we get 'find_group_last (i) == i + 1'
(so, the same).
> In a previous iteration of the refcount overhaul patch, we had the
> "magic" code fragment:
>
>> + for (int j = 0; j < 2; j++)
>> + gomp_map_vars_async (acc_dev, aq,
>> + (j == 0 || pointer == 2) ? 1 : 2,
>> + &hostaddrs[i + j], NULL,
>> + &sizes[i + j], &kinds[i + j], true,
>> + GOMP_MAP_VARS_OPENACC_ENTER_DATA);
> The "pointer == 2" case (i.e. with a GOMP_MAP_TO and a
> GOMP_MAP_POINTER)
So, that's the example given above.
> will also handle the mappings separately in both the
> earlier patch iteration
ACK, given the "previous iteration" code presented above.
> and this one.
NACK? Given 'find_group_last (i) == i + 1', that means that
'GOMP_MAP_TO' and 'GOMP_MAP_POINTER' get mapped as one group?
On the other hand, it still does match the current 'find_pointer'
behavior?
But what should the behavior here be: 'GOMP_MAP_TO', 'GOMP_MAP_POINTER'
each separate, or as one group?
Confusing stuff. :-|
Grüße
Thomas
signature.asc
Description: PGP signature
