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 | 526 | /* Preempt flag, IB should set Pre_enb bit if PREEMPT flag detected */ |
| 527 | 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 | 533 | struct drm_amdgpu_cs_chunk_ib { |
| 530 | 534 | __u32 _pad; |
| 531 | 535 | /** AMDGPU_IB_FLAG_* */ |