original SWITCH_BUFFER scheme for GFX8 doesn't support virtualization, and it give worse performance, now we use the classic scheme to support virtualization and gain more performance by only inserting one SWITCH_BUFFER at the bottom of each command submit, and doing this requires 128dw NOP inserted after VM flush as well to prevent CE trigger vm fault before De vm_flush done.
now 256 dw for each command submit is enough. Monk Liu (2): drm/amdgpu:fix gfx ib schedule drm/amdgpu:256dw is enough for one ib_schedule drivers/gpu/drm/amd/amdgpu/amdgpu.h | 2 ++ drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c | 10 +++++++++- drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c | 28 +++++++++++++--------------- 3 files changed, 24 insertions(+), 16 deletions(-) -- 1.9.1 _______________________________________________ amd-gfx mailing list amd-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/amd-gfx
