fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-01 01:38:06 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 5

brw: switch mesh/task URB fence prior to EOT to GPU

Browse source 
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Cc: mesa-stable
Reviewed-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Rohan Garg <rohan.garg@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/30849>
This commit is contained in:
Lionel Landwerlin 2024-08-26 14:43:58 +03:00 • committed by Marge Bot
parent fda6f8638a
commit 93fba40389
1 changed files with 12 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

13
src/intel/compiler/brw_compile_mesh.cpp
View file

@ -270,7 +270,18 @@ brw_emit_urb_fence(fs_visitor &s)
brw_imm_ud(true),
brw_imm_ud(0));
fence->sfid = BRW_SFID_URB;
fence->desc = lsc_fence_msg_desc(s.devinfo, LSC_FENCE_LOCAL,
/* The logical thing here would likely be a THREADGROUP fence but that's
* still failing some tests like in dEQP-VK.mesh_shader.ext.query.*
*
* Gfx12.5 has a comment about this on BSpec 53533 :
*
* "If fence scope is Local or Threadgroup, HW ignores the flush type
* and operates as if it was set to None (no flush)"
*
* Software workaround from HSD-22014129519 indicates that a GPU fence
* resolves the issue.
*/
fence->desc = lsc_fence_msg_desc(s.devinfo, LSC_FENCE_GPU,
LSC_FLUSH_TYPE_NONE, true);
bld.exec_all().group(1, 0).emit(FS_OPCODE_SCHEDULING_FENCE,