On Mon, 2016-09-26 at 08:42 -0700, Eric Anholt wrote:
> Timothy Arceri <timothy.arc...@collabora.com> writes:
>
> >
> > On Sun, 2016-09-25 at 13:26 -0700, Eric Anholt wrote:
> > >
> > > Timothy Arceri <timothy.arc...@collabora.com> writes:
> > > >
> > > > +static void
> > > > +test_put_key_and_get_key(void)
> > > > +{
> > > > + struct program_cache *cache;
> > > > + bool result;
> > > > +
> > > > + uint8_t key_a[20] =
> > > > { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9,
> > > > + 10, 11, 12, 13, 14, 15, 16, 17, 18,
> > > > 19};
> > > > + uint8_t key_b[20] = { 20, 21, 22, 23, 24, 25, 26, 27, 28,
> > > > 29,
> > > > + 30, 33, 32, 33, 34, 35, 36, 37, 38,
> > > > 39};
> > > > + uint8_t key_a_collide[20] =
> > > > + { 0, 1, 42, 43, 44, 45, 46, 47, 48,
> > > > 49,
> > > > + 50, 55, 52, 53, 54, 55, 56, 57, 58,
> > > > 59};
> > > > +
> > > > + cache = cache_create();
> > > > +
> > > > + /* First test that cache_has_key returns false before
> > > > cache_put_key */
> > > > + result = cache_has_key(cache, key_a);
> > > > + expect_equal(result, 0, "cache_has_key before key added");
> > > > +
> > > > + /* Then a couple of tests of cache_put_key followed by
> > > > cache_has_key */
> > > > + cache_put_key(cache, key_a);
> > > > + result = cache_has_key(cache, key_a);
> > > > + expect_equal(result, 1, "cache_has_key after key added");
> > > > +
> > > > + cache_put_key(cache, key_b);
> > > > + result = cache_has_key(cache, key_b);
> > > > + expect_equal(result, 1, "2nd cache_has_key after key
> > > > added");
> > > > +
> > > > + /* Test that a key with the same two bytes as an existing
> > > > key
> > > > + * forces an eviction.
> > > > + */
> > > > + cache_put_key(cache, key_a_collide);
> > > > + result = cache_has_key(cache, key_a_collide);
> > > > + expect_equal(result, 1, "put_key of a colliding key lands
> > > > in
> > > > the cache");
> > > > +
> > > > + result = cache_has_key(cache, key_a);
> > > > + expect_equal(result, 0, "put_key of a colliding key evicts
> > > > from
> > > > the cache");
> > > > +
> > > > + /* And finally test that we can re-add the original key to
> > > > re-
> > > > evict
> > > > + * the colliding key.
> > > > + */
> > > > + cache_put_key(cache, key_a);
> > > > + result = cache_has_key(cache, key_a);
> > > > + expect_equal(result, 1, "put_key of original key lands
> > > > again");
> > > > +
> > > > + result = cache_has_key(cache, key_a_collide);
> > > > + expect_equal(result, 0, "put_key of oiginal key evicts the
> > > > colliding key");
> > >
> > > "original"
> > >
> > > I haven't yet figured out what the purpose of
> > > cache_put_key()/cache_has_key() are. I suppose I'll find out
> > > later
> > > in
> > > the series.
> >
> > Since we cache a program rather than individual shaders we set a
> > cache
> > key for each shader and opportunistically skip compiling it next
> > time
> > we see the shader.
> >
> > If we happen to be using the shader to create a program we haven't
> > seen
> > before we end up having to fall back to compiling the shader
> > later.
>
> That works out better than just always skipping shader compile until
> link time and you find that you need it?
Good question. I haven't really done any profiling, this is the way
Carl and Kristian come up with doing it before I came along. For the
initial implementation I've been focusing on correctness rather than
speed. I'd really like to leave performance tweaks until after we land
something thats working.
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
