Hi! On Thu, 09 Jan 2014 23:21:26 +0100, I wrote: > On Thu, 9 Jan 2014 21:43:35 +0100, Jakub Jelinek <ja...@redhat.com> wrote: > > On Thu, Jan 09, 2014 at 09:38:25PM +0100, Thomas Schwinge wrote: > > > In gimplify.c:gimplify_adjust_omp_clauses_1, does the case for > > > GOVD_MAP_TO_ONLY have a real current use case (I couldn't spot any), or > > > is it "just for completeness"? > > > > It is typically for any artificial vars that are known not to need copying > > back, such as various artifical vars used for VLAs (say if you do sizeof > > on vla inside of target region, typesizes etc.). The testsuite coverage is > > insufficient here, sure. GOVD_MAP_TO_ONLY is kind of GOVD_FIRSTPRIVATE > > for the target regions, as opposed to GOVD_SHARED. > > Thanks, that makes sense (and I put a TODO item up for adding such tests > to the testsuite), but I still can't manage to find one that actually > triggers the GOVD_MAP_TO_ONLY case in gimplify_adjust_omp_clauses_1. > Maybe it's just too late today.
Haha, and here we go; need something nested to trigger this:
int l = 10;
float c[l];
#pragma omp target map(c[2:4])
{
#pragma omp target
{
int s = sizeof c;
}
}
Grüße,
Thomas
pgpjiq6luXvVW.pgp
Description: PGP signature
