radv: invalidate L2 instead of only writeback L2 when using DCC stores
It seems INV_L2 is the right thing to do, especially for RDNA2 chips with non-coherent RBs (NAVI22 is one of these). This fixes DCC corruption. Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/6476 Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/7507 Cc: mesa-stable Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com> Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl> Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19516>
This commit is contained in:

committed by
Marge Bot

parent
26c8fedc1b
commit
06adf6ad11
@@ -4574,11 +4574,8 @@ radv_src_access_flush(struct radv_cmd_buffer *cmd_buffer, VkAccessFlags2 src_fla
|
||||
}
|
||||
}
|
||||
|
||||
/* This is valid even for the rb_noncoherent_dirty case, because with how we account for
|
||||
* dirtyness, if it isn't dirty it doesn't contain the data at all and hence doesn't need
|
||||
* invalidating. */
|
||||
if (!image_is_coherent)
|
||||
flush_bits |= RADV_CMD_FLAG_WB_L2;
|
||||
flush_bits |= RADV_CMD_FLAG_INV_L2;
|
||||
break;
|
||||
case VK_ACCESS_2_ACCELERATION_STRUCTURE_WRITE_BIT_KHR:
|
||||
case VK_ACCESS_2_TRANSFORM_FEEDBACK_WRITE_BIT_EXT:
|
||||
|
Reference in New Issue
Block a user