On Tue, Oct 30, 2018 at 05:30:46PM +0000, Chris Wilson wrote:
> Quoting Ville Syrjälä (2018-10-30 13:32:39)
> > On Mon, Oct 29, 2018 at 08:49:58PM +0000, Chris Wilson wrote:
> > > Older platforms require fence registers to perform blits, and so
> > > userspace is expected to mark up the objects to request fences be
> > > assigned.
> > >
> > > Fixes: ff2db94acb53 ("igt/gem_tiled_fence_blits: Remove libdrm_intel dependence")
> > > Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=108591
> > > Signed-off-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
> > > ---
> > > tests/i915/gem_tiled_fence_blits.c | 2 ++
> > > 1 file changed, 2 insertions(+)
> > >
> > > diff --git a/tests/i915/gem_tiled_fence_blits.c b/tests/i915/gem_tiled_fence_blits.c
> > > index 7560fa52..e40a7b43 100644
> > > --- a/tests/i915/gem_tiled_fence_blits.c
> > > +++ b/tests/i915/gem_tiled_fence_blits.c
> > > @@ -141,6 +141,8 @@ static void run_test(int fd, int count)
> > >
> > > memset(reloc, 0, sizeof(reloc));
> > > memset(obj, 0, sizeof(obj));
> > > + obj[0].flags = EXEC_OBJECT_NEEDS_FENCE;
> > > + obj[1].flags = EXEC_OBJECT_NEEDS_FENCE;
> >
> > No harm in always asking for the fence I suppose.
>
> If the fence isn't required for the GPU (i.e. gen4+) we just quietly
> pretend it was never requested.
Ah yes. I do recall that being the case. But I couldn't immediately
find the code for it. Now after a second look it seems to be this:
if (!eb->reloc_cache.has_fence) {
entry->flags &= ~EXEC_OBJECT_NEEDS_FENCE;
} else {
...
I guess the "reloc_cache" name threw me off the track the first time
around.
--
Ville Syrjälä
Intel
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
