[Mesa-dev] [PATCH kmscube 3/4] atomic: Duplicate fence before SwapBuffers

[Daniel Stone]

Hi,

On 2 May 2017 at 14:49, Rob Clark <robdclark at gmail.com> wrote:
> I think this is papering over an i965 bug.. in fact doing the
> DupNativeFenceFD could trigger a flush which you don't actually want
> on a tiler.

Well, when it's immediately before SwapBuffers, I don't think it
matters that much. :P

> The corresponding wording in the spec:
>     [...]

Yeah, what you said on IRC makes total sense: if someone wants a
purely internal sync, then they can use EGL_SYNC_FENCE_KHR, and if
they want it then they can use EGL_SYNC_NATIVE_FENCE_ANDROID. So, if
the Intel driver has performance issues (not exposed by kmscube,
presumably) with having to construct a dummy batchbuffer solely to
fence on, then they could instead look at differentiating based on
EGLSync object type.

Dropping both this and 4/4 (which was fixed in between when I last
updated, and rebasing) still works fine on Skylake for me with drm-tip
and ridiculously current Mesa.

Cheers,
Daniel