Comme=
nt # 14
on bug 10234=
9
from Ilia Mirkin
I believe the issue is mostly resolved by:
https://patchwo=
rk.freedesktop.org/patch/270609/
There's a lingering issue, which I don't think was truly being hit, but fix=
ed
by:
https://patchwo=
rk.freedesktop.org/patch/270610/
Of course when I went to test these patches with xonotic, I was immediately
confronted by the fact that we don't handle 3d transfers (which ~never used=
to
happen before, but some recent improvements have made it a thing that happe=
ns a
lot more often). I quickly hacked around that with the below patch, but tha=
t's
obviously not generally acceptable.
diff --git a/src/gallium/auxiliary/util/u_transfer.c
b/src/gallium/auxiliary/util/u_transfer.c
index 3089bcb1f34..3550e1e26d3 100644
--- a/src/gallium/auxiliary/util/u_transfer.c
+++ b/src/gallium/auxiliary/util/u_transfer.c
@@ -56,11 +56,16 @@ void u_default_texture_subdata(struct p=
ipe_context *pipe,
/* texture_subdata implicitly discards the rewritten buffer range */
usage |=3D PIPE_TRANSFER_DISCARD_RANGE;
+ for (int z =3D box->z; z < box->z + box->depth; z++) {
+ struct pipe_box t =3D *box;
+ t.z =3D z;
+ t.depth =3D 1;
+
map =3D pipe->transfer_map(pipe,
resource,
level,
usage,
- box, &transfer);
+ &t, &transfer);
if (!map)
return;
@@ -71,13 +76,14 @@ void u_default_texture_subdata(struct p=
ipe_context *pipe,
0, 0, 0,
box->width,
box->height,
- box->depth,
+ 1,
src_data,
stride, /* bytes */
layer_stride, /* bytes */
- 0, 0, 0);
+ 0, 0, t.z);
pipe_transfer_unmap(pipe, transfer);
+ }
}
FWIW xonotic looks terrible on the nv34 -- all the colors are messed up, and
maybe some of the geometry too. But it's the same with 18.3.