+static unsigned int
+intel_horizontal_texture_alignment_unit(struct brw_context *brw,
+ mesa_format format)
+{
+ /**
+ * From the "Alignment Unit Size" section of various specs, namely:
+ * - Gen3 Spec: "Memory Data Formats" Volume, Section 1.20.1.4
+ * - i965 and G45 PRMs: Volume 1, Section 6.17.3.4.
+ * - Ironlake and Sandybridge PRMs: Volume 1, Part 1, Section 7.18.3.4
+ * - BSpec (for Ivybridge and slight variations in separate stencil)
+ *
+ * +----------------------------------------------------------------------+
+ * | | alignment unit width ("i") |
+ * | Surface Property |-----------------------------|
+ * | | 915 | 965 | ILK | SNB | IVB |
+ * +----------------------------------------------------------------------+
+ * | YUV 4:2:2 format | 8 | 4 | 4 | 4 | 4 |
+ * | BC1-5 compressed format (DXTn/S3TC) | 4 | 4 | 4 | 4 | 4 |
+ * | FXT1 compressed format | 8 | 8 | 8 | 8 | 8 |
+ * | Depth Buffer (16-bit) | 4 | 4 | 4 | 4 | 8 |
+ * | Depth Buffer (other) | 4 | 4 | 4 | 4 | 4 |
+ * | Separate Stencil Buffer | N/A | N/A | 8 | 8 | 8 |
+ * | All Others | 4 | 4 | 4 | 4 | 4 |
+ * +----------------------------------------------------------------------+
+ *
+ * On IVB+, non-special cases can be overridden by setting the SURFACE_STATE
+ * "Surface Horizontal Alignment" field to HALIGN_4 or HALIGN_8.
+ */
+ if (_mesa_is_format_compressed(format)) {
+ /* The hardware alignment requirements for compressed textures
+ * happen to match the block boundaries.
+ */
+ unsigned int i, j;
+ _mesa_get_format_block_size(format, &i, &j);
+ return i;
+ }
+
+ if (format == MESA_FORMAT_S_UINT8)
+ return 8;
+
+ if (brw->gen >= 7 && format == MESA_FORMAT_Z_UNORM16)
+ return 8;
+
+ return 4;
+}
+
+static unsigned int
+intel_vertical_texture_alignment_unit(struct brw_context *brw,
+ mesa_format format, bool multisampled)
+{
+ /**
+ * From the "Alignment Unit Size" section of various specs, namely:
+ * - Gen3 Spec: "Memory Data Formats" Volume, Section 1.20.1.4
+ * - i965 and G45 PRMs: Volume 1, Section 6.17.3.4.
+ * - Ironlake and Sandybridge PRMs: Volume 1, Part 1, Section 7.18.3.4
+ * - BSpec (for Ivybridge and slight variations in separate stencil)
+ *
+ * +----------------------------------------------------------------------+
+ * | | alignment unit height ("j") |
+ * | Surface Property |-----------------------------|
+ * | | 915 | 965 | ILK | SNB | IVB |
+ * +----------------------------------------------------------------------+
+ * | BC1-5 compressed format (DXTn/S3TC) | 4 | 4 | 4 | 4 | 4 |
+ * | FXT1 compressed format | 4 | 4 | 4 | 4 | 4 |
+ * | Depth Buffer | 2 | 2 | 2 | 4 | 4 |
+ * | Separate Stencil Buffer | N/A | N/A | N/A | 4 | 8 |
+ * | Multisampled (4x or 8x) render target | N/A | N/A | N/A | 4 | 4 |
+ * | All Others | 2 | 2 | 2 | * | * |
+ * +----------------------------------------------------------------------+
+ *
+ * Where "*" means either VALIGN_2 or VALIGN_4 depending on the setting of
+ * the SURFACE_STATE "Surface Vertical Alignment" field.
+ */
+ if (_mesa_is_format_compressed(format))
+ return 4;
+
+ if (format == MESA_FORMAT_S_UINT8)
+ return brw->gen >= 7 ? 8 : 4;
+
+ /* Broadwell only supports VALIGN of 4, 8, and 16. The BSpec says 4
+ * should always be used, except for stencil buffers, which should be 8.
+ */
+ if (brw->gen >= 8)
+ return 4;
+
+ if (multisampled)
+ return 4;
+
+ GLenum base_format = _mesa_get_format_base_format(format);
+
+ if (brw->gen >= 6 &&
+ (base_format == GL_DEPTH_COMPONENT ||
+ base_format == GL_DEPTH_STENCIL)) {
+ return 4;
+ }
+
+ if (brw->gen == 7) {
+ /* On Gen7, we prefer a vertical alignment of 4 when possible, because
+ * that allows Y tiled render targets.
+ *
+ * From the Ivy Bridge PRM, Vol4 Part1 2.12.2.1 (SURFACE_STATE for most
+ * messages), on p64, under the heading "Surface Vertical Alignment":
+ *
+ * Value of 1 [VALIGN_4] is not supported for format YCRCB_NORMAL
+ * (0x182), YCRCB_SWAPUVY (0x183), YCRCB_SWAPUV (0x18f), YCRCB_SWAPY
+ * (0x190)
+ *
+ * VALIGN_4 is not supported for surface format R32G32B32_FLOAT.
+ */
+ if (base_format == GL_YCBCR_MESA || format == MESA_FORMAT_RGB_FLOAT32)
+ return 2;
+
+ return 4;
+ }
+
+ return 2;
+}
+
+static void
+brw_miptree_layout_2d(struct intel_mipmap_tree *mt)
+{
+ unsigned x = 0;
+ unsigned y = 0;
+ unsigned width = mt->physical_width0;
+ unsigned height = mt->physical_height0;
+ unsigned depth = mt->physical_depth0; /* number of array layers. */
+
+ mt->total_width = mt->physical_width0;
+
+ if (mt->compressed) {
+ mt->total_width = ALIGN(mt->physical_width0, mt->align_w);
+ }
+
+ /* May need to adjust width to accomodate the placement of
+ * the 2nd mipmap. This occurs when the alignment
+ * constraints of mipmap placement push the right edge of the
+ * 2nd mipmap out past the width of its parent.
+ */
+ if (mt->first_level != mt->last_level) {
+ unsigned mip1_width;
+
+ if (mt->compressed) {
+ mip1_width = ALIGN(minify(mt->physical_width0, 1), mt->align_w) +
+ ALIGN(minify(mt->physical_width0, 2), mt->align_w);
+ } else {
+ mip1_width = ALIGN(minify(mt->physical_width0, 1), mt->align_w) +
+ minify(mt->physical_width0, 2);
+ }
+
+ if (mip1_width > mt->total_width) {
+ mt->total_width = mip1_width;
+ }
+ }
+
+ mt->total_height = 0;
+
+ for (unsigned level = mt->first_level; level <= mt->last_level; level++) {
+ unsigned img_height;
+
+ intel_miptree_set_level_info(mt, level, x, y, depth);
+
+ img_height = ALIGN(height, mt->align_h);
+ if (mt->compressed)
+ img_height /= mt->align_h;
+
+ if (mt->array_layout == ALL_SLICES_AT_EACH_LOD) {
+ /* Compact arrays with separated miplevels */
+ img_height *= depth;
+ }
+
+ /* Because the images are packed better, the final offset
+ * might not be the maximal one:
+ */
+ mt->total_height = MAX2(mt->total_height, y + img_height);
+
+ /* Layout_below: step right after second mipmap.
+ */
+ if (level == mt->first_level + 1) {
+ x += ALIGN(width, mt->align_w);
+ } else {
+ y += img_height;
+ }
+
+ width = minify(width, 1);
+ height = minify(height, 1);
+ }
+}
+
+static void
+align_cube(struct intel_mipmap_tree *mt)
+{
+ /* The 965's sampler lays cachelines out according to how accesses
+ * in the texture surfaces run, so they may be "vertical" through
+ * memory. As a result, the docs say in Surface Padding Requirements:
+ * Sampling Engine Surfaces that two extra rows of padding are required.
+ */
+ if (mt->target == GL_TEXTURE_CUBE_MAP)
+ mt->total_height += 2;
+}
+