[Intel-gfx] [PATCH igt v2] igt/gem_linear_blits: Compute GTT size using 4G limit
Arkadiusz Hiler
arkadiusz.hiler at intel.com
Tue Dec 19 10:59:45 UTC 2017
On Fri, Dec 15, 2017 at 12:22:01PM +0000, Chris Wilson wrote:
> Quoting Chris Wilson (2017-09-13 11:39:14)
> > Both gem_linear_blits and gem_tiled_blit do not request the full 48b
> > GTT layout for their objects, restricting themselves to 4G. The
> > underlying test that they trigger eviction is unaffected by this
> > restriction, so we can simply reduce their memory requirements to fill
> > the low 4G GTT space and so allow them to run on 48b machines.
> >
> > v2: gem_tiled_fenced_blits as well
> >
> > Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
>
> Ping ?
Pong.
Requeued this for testing, as the old results seem to be lost in time
and space.
> > ---
> > tests/gem_linear_blits.c | 18 ++++++++++++++----
> > tests/gem_tiled_blits.c | 16 ++++++++++++----
> > tests/gem_tiled_fence_blits.c | 11 +++++++++--
> > 3 files changed, 35 insertions(+), 10 deletions(-)
> >
> > diff --git a/tests/gem_linear_blits.c b/tests/gem_linear_blits.c
> > index eccfbcdd..8297416c 100644
> > --- a/tests/gem_linear_blits.c
> > +++ b/tests/gem_linear_blits.c
> > @@ -215,6 +215,8 @@ static void run_test(int fd, int count)
> > free(handle);
> > }
> >
> > +#define MAX_32b ((1ull << 32) - 4096)
> > +
> > int main(int argc, char **argv)
> > {
> > int fd = 0;
> > @@ -230,20 +232,28 @@ int main(int argc, char **argv)
> > run_test(fd, 2);
> >
> > igt_subtest("normal") {
> > - int count;
> > + uint64_t count;
> >
> > - count = 3 * gem_aperture_size(fd) / (1024*1024) / 2;
> > + count = gem_aperture_size(fd);
> > + if (count >> 32)
> > + count = MAX_32b;
> > + count = 3 * count / (1024*1024) / 2;
Having that in 4 places may justify an introduction of a helper.
Nonetheless,
Reviewed-by: Arkadiusz Hiler <arkadiusz.hiler at intel.com>
- Arek
> > igt_require(count > 1);
> > intel_require_memory(count, sizeof(linear), CHECK_RAM);
> > +
> > run_test(fd, count);
> > }
More information about the Intel-gfx
mailing list