load z because we are cool like that

src/amd/common/ac_nir_lower_ngg.c

@@ -1124,6 +1124,7 @@ add_deferred_attribute_culling(nir_builder *b, nir_cf_list *original_extracted_c
             nir_ssa_def *xy = nir_build_load_shared(b, 2, 32, vtx_addr[vtx], .align_mul = 4u, .base = lds_es_pos_x);
             pos[vtx][0] = nir_channel(b, xy, 0);
             pos[vtx][1] = nir_channel(b, xy, 1);
+            pos[vtx][2] = nir_channel(b, xy, 2);
          }
 
          /* See if the current primitive is accepted */