public inbox for intel-xe@lists.freedesktop.org
 help / color / mirror / Atom feed
* [PATCH] drm/xe: Do not blindly copy system CCS pages when shrinking
@ 2026-04-30 19:09 Matthew Brost
  2026-04-30 19:16 ` ✓ CI.KUnit: success for " Patchwork
  2026-04-30 20:05 ` ✗ Xe.CI.BAT: failure " Patchwork
  0 siblings, 2 replies; 3+ messages in thread
From: Matthew Brost @ 2026-04-30 19:09 UTC (permalink / raw)
  To: intel-xe, dri-devel

If the PAT index for a BO is configured without compression, copying CCS
pages during BO shrinking provides no value. Skip CCS copies for BOs that
have never been configured with compressed PAT indexes.

Once a BO is configured with a compressed PAT index, this becomes a
one-way transition that enables CCS copies. This heuristic is imprecise,
but simple and functional.

Signed-off-by: Matthew Brost <matthew.brost@intel.com>

---

When running 10x tabs webgl aquarium, it was observed a ratio of ~6.5 to
1 of BOs that can possible enable compression (xe_bo_needs_ccs_pages
returns true) actually enable compression in the PAT index and overall
CCS copies are reduced by roughly 1/2 when the shrinker is triggered.
---
 drivers/gpu/drm/xe/xe_bo.c       | 2 ++
 drivers/gpu/drm/xe/xe_bo_types.h | 3 +++
 drivers/gpu/drm/xe/xe_vm.c       | 3 +++
 3 files changed, 8 insertions(+)

diff --git a/drivers/gpu/drm/xe/xe_bo.c b/drivers/gpu/drm/xe/xe_bo.c
index 5ce60d161e09..3fbb82ece4de 100644
--- a/drivers/gpu/drm/xe/xe_bo.c
+++ b/drivers/gpu/drm/xe/xe_bo.c
@@ -1105,6 +1105,8 @@ static int xe_bo_move(struct ttm_buffer_object *ttm_bo, bool evict,
 			flags |= XE_MIGRATE_CLEAR_FLAG_CCS_DATA;
 
 		fence = xe_migrate_clear(migrate, bo, new_mem, flags);
+	} else if (handle_system_ccs && !bo->ccs_used) {
+		fence = dma_fence_get_stub();
 	} else {
 		fence = xe_migrate_copy(migrate, bo, bo, old_mem, new_mem,
 					handle_system_ccs);
diff --git a/drivers/gpu/drm/xe/xe_bo_types.h b/drivers/gpu/drm/xe/xe_bo_types.h
index 9c199badd9b2..76f73a3bf2ca 100644
--- a/drivers/gpu/drm/xe/xe_bo_types.h
+++ b/drivers/gpu/drm/xe/xe_bo_types.h
@@ -88,6 +88,9 @@ struct xe_bo {
 	/** @ccs_cleared: true means that CCS region of BO is already cleared */
 	bool ccs_cleared;
 
+	/** @ccs_used: true means that CCS region of BO is used */
+	bool ccs_used;
+
 	/** @bb_ccs: BB instructions of CCS read/write. Valid only for VF */
 	struct xe_mem_pool_node *bb_ccs[XE_SRIOV_VF_CCS_CTX_COUNT];
 
diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c
index c3836f6eab35..638037e71e2d 100644
--- a/drivers/gpu/drm/xe/xe_vm.c
+++ b/drivers/gpu/drm/xe/xe_vm.c
@@ -3054,6 +3054,9 @@ static int vma_lock_and_validate(struct drm_exec *exec, struct xe_vma *vma,
 
 		if (flags.request_decompress)
 			err = xe_bo_decompress(bo);
+
+		if (xe_pat_index_get_comp_en(vm->xe, vma->attr.pat_index))
+			bo->ccs_used = true;
 	}
 
 	return err;
-- 
2.34.1


^ permalink raw reply related	[flat|nested] 3+ messages in thread
end of thread, other threads:[~2026-04-30 20:05 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-04-30 19:09 [PATCH] drm/xe: Do not blindly copy system CCS pages when shrinking Matthew Brost
2026-04-30 19:16 ` ✓ CI.KUnit: success for " Patchwork
2026-04-30 20:05 ` ✗ Xe.CI.BAT: failure " Patchwork

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox