答复: [PATCH] drm/amdgpu:guarantee 128dws between vm flush and IB(v3)

2017-01-21 Thread Liu, Monk
> 代表 Liu, Monk <monk@amd.com> 发送时间: 2017年1月22日 10:47:03 收件人: Christian König; amd-gfx@lists.freedesktop.org 主题: 答复: [PATCH] drm/amdgpu:guarantee 128dws between vm flush and IB(v3) 1, windows separate vm flush and gfx submit, so if with vm flush windows actually submit 256 dw totally

答复: [PATCH] drm/amdgpu:guarantee 128dws between vm flush and IB(v3)

2017-01-21 Thread Liu, Monk
1, windows separate vm flush and gfx submit, so if with vm flush windows actually submit 256 dw totally (128 for vm flush and 128 for gfx submit) 2, for gfx submit frame: windows always emit pipeline sync(name is "w0ait_3d_idle" for windows) 3, for vm flush frame: windows always emit pipeline

Re: 答复: 答复: [PATCH] drm/amdgpu:guarantee 128dws between vm flush and IB

2017-01-19 Thread Christian König
*发件人:* amd-gfx <amd-gfx-boun...@lists.freedesktop.org> 代表 Christian König <deathsim...@vodafone.de> *发送时间:* 2017年1月18日 20:52:14 *收件人:* Liu, Monk; amd-gfx@lists.freedesktop.org *主题:* Re: 答复: [PATCH] drm/amdgpu:guarantee 128dws between vm f

答复: 答复: [PATCH] drm/amdgpu:guarantee 128dws between vm flush and IB

2017-01-18 Thread Liu, Monk
发件人: amd-gfx <amd-gfx-boun...@lists.freedesktop.org> 代表 Christian König <deathsim...@vodafone.de> 发送时间: 2017年1月18日 20:52:14 收件人: Liu, Monk; amd-gfx@lists.freedesktop.org 主题: Re: 答复: [PATCH] drm/amdgpu:guarantee 128dws between vm flush and I

Re: 答复: [PATCH] drm/amdgpu:guarantee 128dws between vm flush and IB

2017-01-18 Thread Christian König
suppose hw_submit_count in gpu_scheduler is 2 (by default), and without suppress frame size under 256, we may sometime submit 256dw, and sometime submit 512 dw, and that could lead to CPU override ring buffer content which is under processing by GPU, e.g. : The max_dw parameter for

答复: [PATCH] drm/amdgpu:guarantee 128dws between vm flush and IB

2017-01-18 Thread Liu, Monk
>First question why do we want to limit the dw per submit to 256? Some >limitation for SRIOV? [ML] suppose hw_submit_count in gpu_scheduler is 2 (by default), and without suppress frame size under 256, we may sometime submit 256dw, and sometime submit 512 dw, and that could lead to CPU