intel/isl: Reduce miptail slot usage to allow CCS

We currently disable CCS if a surface uses more than 11 slots in a
miptail. However, ISL generally configures surfaces to be compatible
with compression. For consistency, reduce the number of slots used in
miptails in order to allow compression.

If drivers prefer to have a more compact layout, they can pass the
ISL_SURF_USAGE_DISABLE_AUX_BIT flag at surface creation time.

Reviewed-by: Rohan Garg <rohan.garg@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/30081>
This commit is contained in:
Nanley Chery
2024-04-16 15:07:52 -04:00
committed by Marge Bot
parent 5af81ed7c5
commit 19ed0e1685

View File

@@ -1749,6 +1749,19 @@ isl_choose_miptail_start_level(const struct isl_device *dev,
uint32_t max_miptail_levels = tile_info->max_miptail_levels;
if (max_miptail_levels > 11 &&
_isl_surf_info_supports_ccs(dev, info->format, info->usage)) {
/* SKL PRMs, Volume 5: Memory Views, Tiling and Mip Tails for 2D
* Surfaces:
*
* "Lossless compression must not be used on surfaces which have MIP
* Tail which contains MIPs for Slots greater than 11."
*
* Reduce the slot consumption to keep compression enabled.
*/
max_miptail_levels = 11;
}
/* Start with the minimum number of levels that will fit in the tile */
uint32_t min_miptail_start =
info->levels > max_miptail_levels ? info->levels - max_miptail_levels : 0;
@@ -3093,20 +3106,6 @@ isl_surf_supports_ccs(const struct isl_device *dev,
if (ISL_GFX_VER(dev) >= 9 && surf->tiling == ISL_TILING_X)
return false;
/* SKL PRMs, Volume 5: Memory Views, Tiling and Mip Tails for 2D Surfaces:
*
* "Lossless compression must not be used on surfaces which have MIP
* Tail which contains MIPs for Slots greater than 11."
*/
if (surf->miptail_start_level < surf->levels) {
const uint32_t miptail_levels = surf->levels - surf->miptail_start_level;
if (miptail_levels + isl_get_miptail_base_row(surf->tiling) > 11) {
assert(isl_tiling_is_64(surf->tiling) ||
isl_tiling_is_std_y(surf->tiling));
return false;
}
}
/* From the workarounds section in the SKL PRM:
*
* "RCC cacheline is composed of X-adjacent 64B fragments instead of