[PATCH 4/5] drm/amdgpu: use bulk moves for efficient VM LRU handling
Huang Rui
ray.huang at amd.com
Fri Aug 10 09:03:05 UTC 2018
On Thu, Aug 09, 2018 at 08:27:10PM +0800, Koenig, Christian wrote:
> Am 09.08.2018 um 14:25 schrieb Huang Rui:
> > On Thu, Aug 09, 2018 at 03:18:55PM +0800, Koenig, Christian wrote:
> >> Am 09.08.2018 um 08:18 schrieb Huang Rui:
> >>> On Wed, Aug 08, 2018 at 06:47:49PM +0800, Christian König wrote:
> >>>> Am 08.08.2018 um 11:59 schrieb Huang Rui:
> >>>>> I continue to work for bulk moving that based on the proposal by Christian.
> >>>>>
> >>>>> Background:
> >>>>> amdgpu driver will move all PD/PT and PerVM BOs into idle list. Then move all of
> >>>>> them on the end of LRU list one by one. Thus, that cause so many BOs moved to
> >>>>> the end of the LRU, and impact performance seriously.
> >>>>>
> >>>>> Then Christian provided a workaround to not move PD/PT BOs on LRU with below
> >>>>> patch:
> >>>>> "drm/amdgpu: band aid validating VM PTs"
> >>>>> Commit 0bbf32026cf5ba41e9922b30e26e1bed1ecd38ae
> >>>>>
> >>>>> However, the final solution should bulk move all PD/PT and PerVM BOs on the LRU
> >>>>> instead of one by one.
> >>>>>
> >>>>> Whenever amdgpu_vm_validate_pt_bos() is called and we have BOs which need to be
> >>>>> validated we move all BOs together to the end of the LRU without dropping the
> >>>>> lock for the LRU.
> >>>>>
> >>>>> While doing so we note the beginning and end of this block in the LRU list.
> >>>>>
> >>>>> Now when amdgpu_vm_validate_pt_bos() is called and we don't have anything to do,
> >>>>> we don't move every BO one by one, but instead cut the LRU list into pieces so
> >>>>> that we bulk move everything to the end in just one operation.
> >>>>>
> >>>>> Test data:
> >>>>> +--------------+-----------------+-----------+---------------------------------------+
> >>>>> | |The Talos |Clpeak(OCL)|BusSpeedReadback(OCL) |
> >>>>> | |Principle(Vulkan)| | |
> >>>>> +------------------------------------------------------------------------------------+
> >>>>> | | | |0.319 ms(1k) 0.314 ms(2K) 0.308 ms(4K) |
> >>>>> | Original | 147.7 FPS | 76.86 us |0.307 ms(8K) 0.310 ms(16K) |
> >>>>> +------------------------------------------------------------------------------------+
> >>>>> | Orignial + WA| | |0.254 ms(1K) 0.241 ms(2K) |
> >>>>> |(don't move | 162.1 FPS | 42.15 us |0.230 ms(4K) 0.223 ms(8K) 0.204 ms(16K)|
> >>>>> |PT BOs on LRU)| | | |
> >>>>> +------------------------------------------------------------------------------------+
> >>>>> | Bulk move | 163.1 FPS | 40.52 us |0.244 ms(1K) 0.252 ms(2K) 0.213 ms(4K) |
> >>>>> | | | |0.214 ms(8K) 0.225 ms(16K) |
> >>>>> +--------------+-----------------+-----------+---------------------------------------+
> >>>>>
> >>>>> After test them with above three benchmarks include vulkan and opencl. We can
> >>>>> see the visible improvement than original, and even better than original with
> >>>>> workaround.
> >>>>>
> >>>>> Signed-off-by: Christian König <christian.koenig at amd.com>
> >>>>> Signed-off-by: Huang Rui <ray.huang at amd.com>
> >>>>> ---
> >>>>> drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 22 ++++++++++++++--------
> >>>>> drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 4 ++++
> >>>>> 2 files changed, 18 insertions(+), 8 deletions(-)
> >>>>>
> >>>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
> >>>>> index 9c84770..eda0bb9 100644
> >>>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
> >>>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
> >>>>> @@ -286,6 +286,7 @@ int amdgpu_vm_validate_pt_bos(struct amdgpu_device *adev, struct amdgpu_vm *vm,
> >>>>> {
> >>>>> struct ttm_bo_global *glob = adev->mman.bdev.glob;
> >>>>> struct amdgpu_vm_bo_base *bo_base, *tmp;
> >>>>> + bool validated = false;
> >>>>> int r = 0;
> >>>>>
> >>>>> list_for_each_entry_safe(bo_base, tmp, &vm->evicted, vm_status) {
> >>>>> @@ -295,14 +296,9 @@ int amdgpu_vm_validate_pt_bos(struct amdgpu_device *adev, struct amdgpu_vm *vm,
> >>>>> r = validate(param, bo);
> >>>>> if (r)
> >>>>> break;
> >>>>> -
> >>>>> - spin_lock(&glob->lru_lock);
> >>>>> - ttm_bo_move_to_lru_tail(&bo->tbo, NULL);
> >>>>> - if (bo->shadow)
> >>>>> - ttm_bo_move_to_lru_tail(&bo->shadow->tbo, NULL);
> >>>>> - spin_unlock(&glob->lru_lock);
> >>>>> }
> >>>>>
> >>>>> + validated = true;
> >>>>> if (bo->tbo.type != ttm_bo_type_kernel) {
> >>>>> spin_lock(&vm->moved_lock);
> >>>>> list_move(&bo_base->vm_status, &vm->moved);
> >>>>> @@ -312,6 +308,15 @@ int amdgpu_vm_validate_pt_bos(struct amdgpu_device *adev, struct amdgpu_vm *vm,
> >>>>> }
> >>>>> }
> >>>>>
> >>>>> + if (!validated) {
> >>>>> + spin_lock(&glob->lru_lock);
> >>>>> + ttm_bo_bulk_move_lru_tail(&vm->lru_bulk_move);
> >>>>> + spin_unlock(&glob->lru_lock);
> >>>>> + return 0;
> >>>>> + }
> >>>>> +
> >>>>> + memset(&vm->lru_bulk_move, 0, sizeof(vm->lru_bulk_move));
> >>>>> +
> >>>>> spin_lock(&glob->lru_lock);
> >>>>> list_for_each_entry(bo_base, &vm->idle, vm_status) {
> >>>>> struct amdgpu_bo *bo = bo_base->bo;
> >>>>> @@ -319,9 +324,10 @@ int amdgpu_vm_validate_pt_bos(struct amdgpu_device *adev, struct amdgpu_vm *vm,
> >>>>> if (!bo->parent)
> >>>>> continue;
> >>>>>
> >>>>> - ttm_bo_move_to_lru_tail(&bo->tbo, NULL);
> >>>>> + ttm_bo_move_to_lru_tail(&bo->tbo, &vm->lru_bulk_move);
> >>>>> if (bo->shadow)
> >>>>> - ttm_bo_move_to_lru_tail(&bo->shadow->tbo, NULL);
> >>>>> + ttm_bo_move_to_lru_tail(&bo->shadow->tbo,
> >>>>> + &vm->lru_bulk_move);
> >>>> Here is still one problem: The BOs currently evicted are not yet in the
> >>>> idle state.
> >>>>
> >>>> So this must be moved to the end of the state machine, probably best if
> >>>> we put that into a separate function.
> >>>>
> >>> Thanks, so the evicted BOs should be behind the BOs in idle list that in
> >>> the group between (first, last) of bulk_move_pos, right?
> >> The order doesn't matter.
> >>
> >>> spin_lock(&glob->lru_lock);
> >>> list_for_each_entry(bo_base, &vm->idle, vm_status) {
> >>> struct amdgpu_bo *bo = bo_base->bo;
> >>>
> >>> if (!bo->parent)
> >>> continue;
> >>>
> >>> ttm_bo_move_to_lru_tail(&bo->tbo, &vm->lru_bulk_move);
> >>> if (bo->shadow)
> >>> ttm_bo_move_to_lru_tail(&bo->shadow->tbo,
> >>> &vm->lru_bulk_move);
> >>> }
> >>> list_for_each_entry_safe(bo_base, tmp, &vm->evicted, vm_status) {
> >> That won't work. The BOs get removed from the evicted list when they are
> >> validated.
> >>
> >> You need to walk both the relocated as well as the moved list here.
> >>
> > Is that because, at the start of the function, we already walk the evicted
> > list, and some of BOs are validated and move to moved or relocated list, so
> > here we need also walk both of them to make them together for bulk move?
>
> Yes, exactly.
>
> Alternatively you can add a new function which is called after command
> submission is done to move everything on the LRU.
>
> At this point all BOs should now be on the idle list again and you only
> need to walk that one.
Do you mean we move all BOs to idle list again in amdgpu_vm_bo_update of
amdgpu_cs_ib_vm_chunk subsequently?
But at that time, I see if the BO is evicted, it will be moved to evicted
list, not to idle list again. Am I missed something?
Thanks,
Ray
>
> Regards,
> Christian.
>
> >
> > Thanks,
> > Ray
> >
> >> Christian.
> >>
> >>> struct amdgpu_bo *bo = bo_base->bo;
> >>> if (!bo)
> >>> continue;
> >>>
> >>> ttm_bo_move_to_lru_tail(&bo->tbo, &vm->lru_bulk_move);
> >>> if (bo->shadow)
> >>> ttm_bo_move_to_lru_tail(&bo->shadow->tbo,
> >>> &vm->lru_bulk_move);
> >>> }
> >>> spin_unlock(&glob->lru_lock);
> >>>
> >>> Thanks,
> >>> Ray
> >>>
> >>>> Regards,
> >>>> Christian.
> >>>>
> >>>>> }
> >>>>> spin_unlock(&glob->lru_lock);
> >>>>>
> >>>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
> >>>>> index 67a15d4..92725ac 100644
> >>>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
> >>>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
> >>>>> @@ -29,6 +29,7 @@
> >>>>> #include <linux/rbtree.h>
> >>>>> #include <drm/gpu_scheduler.h>
> >>>>> #include <drm/drm_file.h>
> >>>>> +#include <drm/ttm/ttm_bo_driver.h>
> >>>>>
> >>>>> #include "amdgpu_sync.h"
> >>>>> #include "amdgpu_ring.h"
> >>>>> @@ -226,6 +227,9 @@ struct amdgpu_vm {
> >>>>>
> >>>>> /* Some basic info about the task */
> >>>>> struct amdgpu_task_info task_info;
> >>>>> +
> >>>>> + /* Store positions of group of BOs */
> >>>>> + struct ttm_lru_bulk_move lru_bulk_move;
> >>>>> };
> >>>>>
> >>>>> struct amdgpu_vm_manager {
>
More information about the dri-devel
mailing list