There is a new IB flag that enables this new behavior. Full invalidation is unnecessary for RELEASE_MEM and doesn't make sense when draw calls from two adjacent gfx IBs run in parallel. This will be the new default for Mesa. v2: bump the version
Showing the changes to include/uapi (1 of 8 files). Show all
Unified · Split
| @@ -526,6 +526,10 @@ union drm_amdgpu_cs { · 525 unmodified lines | |||
| 526 | /* Preempt flag, IB should set Pre_enb bit if PREEMPT flag detected */ | 526 | /* Preempt flag, IB should set Pre_enb bit if PREEMPT flag detected */ |
| 527 | #define AMDGPU_IB_FLAG_PREEMPT (1<<2) | 527 | #define AMDGPU_IB_FLAG_PREEMPT (1<<2) |
| 528 | 528 | ||
| 529 | /* The IB fence should do the L2 writeback but not invalidate any shader | ||
| 530 | * caches (L2/vL1/sL1/I$). */ | ||
| 531 | #define AMDGPU_IB_FLAG_TC_WB_NOT_INVALIDATE (1 << 3) | ||
| 532 | |||
| 529 | struct drm_amdgpu_cs_chunk_ib { | 533 | struct drm_amdgpu_cs_chunk_ib { |
| 530 | __u32 _pad; | 534 | __u32 _pad; |
| 531 | /** AMDGPU_IB_FLAG_* */ | 535 | /** AMDGPU_IB_FLAG_* */ |