[Mesa-dev] [PATCH v3] i965/skl: Fix the qpitch value

[Neil Roberts]

On Skylake the qpitch value is uploaded as part of the surface state
so we don't need to add the extra rows that are done for other
generations. However for 3D textures it needs to be aligned to the
tile height and for depth/stencil textures it needs to be a multiple
of 8. Unlike previous generations the qpitch is measured as a multiple
of the block size for compressed surfaces. When the horizontal mipmap
layout is used for 1D textures then the qpitch is measured in pixels
instead of rows.

v2: Align the depth/stencil textures to a multiple of 8
v3: Add an assert that ALL_SLICES_AT_EACH_LOD is not used. Ignore the
    vertical alignment when picking the qpitch for 1D_ARRAY textures.

---

Here's a v3 which adds an assert about never using
ALL_SLICES_AT_EACH_LOAD. I've also squashed in the patch about
ignoring the vertical alignment for 1D_ARRAY textures.

 src/mesa/drivers/dri/i965/brw_tex_layout.c    | 62 ++++++++++++++++++++++-----
 src/mesa/drivers/dri/i965/intel_mipmap_tree.h | 10 +++--
 2 files changed, 59 insertions(+), 13 deletions(-)

diff --git a/src/mesa/drivers/dri/i965/brw_tex_layout.c b/src/mesa/drivers/dri/i965/brw_tex_layout.c
index 5e70cd2..a67bdd6 100644
--- a/src/mesa/drivers/dri/i965/brw_tex_layout.c
+++ b/src/mesa/drivers/dri/i965/brw_tex_layout.c
@@ -300,24 +300,66 @@ static void
 brw_miptree_layout_texture_array(struct brw_context *brw,
 				 struct intel_mipmap_tree *mt)
 {
-   int h0, h1;
    unsigned height = mt->physical_height0;
    bool layout_1d = use_linear_1d_layout(brw, mt);
-
-   h0 = ALIGN(mt->physical_height0, mt->align_h);
-   h1 = ALIGN(minify(mt->physical_height0, 1), mt->align_h);
-   if (mt->array_layout == ALL_SLICES_AT_EACH_LOD)
-      mt->qpitch = h0;
-   else
-      mt->qpitch = (h0 + h1 + (brw->gen >= 7 ? 12 : 11) * mt->align_h);
-
-   int physical_qpitch = mt->compressed ? mt->qpitch / 4 : mt->qpitch;
+   int physical_qpitch;
 
    if (layout_1d)
       gen9_miptree_layout_1d(mt);
    else
       brw_miptree_layout_2d(mt);
 
+   if (layout_1d) {
+      physical_qpitch = 1;
+      /* When using the horizontal layout the qpitch specifies the distance in
+       * pixels between array slices. The total_width is forced to be a
+       * multiple of the horizontal alignment in brw_miptree_layout_1d (in
+       * this case it's always 64). The vertical alignment is ignored.
+       */
+      mt->qpitch = mt->total_width;
+   } else if (brw->gen >= 9) {
+      GLenum base_format;
+
+      /* ALL_SLICES_AT_EACH_LOD isn't supported on Gen8+ but this code will
+       * effectively end up with a packed qpitch anyway whenever
+       * mt->first_level == mt->last_level.
+       */
+      assert(mt->msaa_layout != ALL_SLICES_AT_EACH_LOD);
+
+      /* On Gen9 we can pick whatever qpitch we like as long as it's aligned
+       * to the vertical alignment so we don't need to add any extra rows.
+       */
+      mt->qpitch = mt->total_height;
+
+      /* If the surface might be used as a stencil buffer or HiZ buffer then
+       * it needs to be a multiple of 8.
+       */
+      base_format = _mesa_get_format_base_format(mt->format);
+      if (_mesa_is_depth_or_stencil_format(base_format))
+         mt->qpitch = ALIGN(mt->qpitch, 8);
+
+      /* 3D textures need to be aligned to the tile height. At this point we
+       * don't know which tiling will be used so let's just align it to 32
+       */
+      if (mt->target == GL_TEXTURE_3D)
+         mt->qpitch = ALIGN(mt->qpitch, 32);
+
+      /* Unlike previous generations the qpitch is now a multiple of the
+       * compressed block size so physical_qpitch matches mt->qpitch.
+       */
+      physical_qpitch = mt->qpitch;
+   } else {
+      int h0 = ALIGN(mt->physical_height0, mt->align_h);
+      int h1 = ALIGN(minify(mt->physical_height0, 1), mt->align_h);
+
+      if (mt->array_layout == ALL_SLICES_AT_EACH_LOD)
+         mt->qpitch = h0;
+      else
+         mt->qpitch = (h0 + h1 + (brw->gen >= 7 ? 12 : 11) * mt->align_h);
+
+      physical_qpitch = mt->compressed ? mt->qpitch / 4 : mt->qpitch;
+   }
+
    for (unsigned level = mt->first_level; level <= mt->last_level; level++) {
       unsigned img_height;
       img_height = ALIGN(height, mt->align_h);
diff --git a/src/mesa/drivers/dri/i965/intel_mipmap_tree.h b/src/mesa/drivers/dri/i965/intel_mipmap_tree.h
index 41b6036..158f0eb 100644
--- a/src/mesa/drivers/dri/i965/intel_mipmap_tree.h
+++ b/src/mesa/drivers/dri/i965/intel_mipmap_tree.h
@@ -403,10 +403,14 @@ struct intel_mipmap_tree
    enum miptree_array_layout array_layout;
 
    /**
-    * The distance in rows between array slices in an uncompressed surface.
+    * The distance in between array slices.
     *
-    * For compressed surfaces, slices are stored closer together physically;
-    * the real distance is (qpitch / block height).
+    * The value is the one that is sent in the surface state. The actual
+    * meaning depends on certain criteria. Usually it is simply the number of
+    * uncompressed rows between each slice. However on Gen9+ for compressed
+    * surfaces it is the number of blocks. For 1D array surfaces that have the
+    * mipmap tree stored horizontally it is the number of pixels between each
+    * slice.
     */
    uint32_t qpitch;
 
-- 
1.9.3