On Fri, 26 Jul 2019 13:31:47 +1200 Barry Song <21cn...@gmail.com> wrote:
> Pekka Paalanen <ppaala...@gmail.com> 于2019年7月26日周五 上午3:30写道： > > Hi, > > > > I don't think that would work too well, because eglSwapBuffers has > > not been called yet at that time, and using glReadPixels would hurt > > performance a lot. > > > > You don't need to use vaapi-recorder per se. You just hook up your > > sink exactly like vaapi-recorder does, and pay attention to > > synchronization. > > well. i'd like to continue in this way. > another possible solution is using a way like ffpmeg kmsgrab: > https://www.ffmpeg.org/ffmpeg-devices.html#Options-10 > > it is using get the DRM framebuffer outside the weston: > https://github.com/FFmpeg/FFmpeg/blob/master/libavdevice/kmsgrab.c > by code like: > static int kmsgrab_read_packet(AVFormatContext *avctx, AVPacket *pkt) > { > ... > plane = drmModeGetPlane(ctx->hwctx->fd, ctx->plane_id); > ... > fb = drmModeGetFB(ctx->hwctx->fd, plane->fb_id); > ... > err = drmPrimeHandleToFD(ctx->hwctx->fd, fb->handle, O_RDONLY, &fd); > ... > desc = av_mallocz(sizeof(*desc)); > ... > } > > do you think if kmsgrab will have some serial synchronization or > performance issues? Hi, first, I am surprised to hear kmsgrab exists or can even work at all. I guess the superuser is allowed to do all kinds of nasty things. I think synchronization issues are unavoidable with kmsgrab. There is no way to know when Weston flips the FB, right? That means you will be polling and randomly sampling the KMS state, so you don't know what frames you'll be getting. I imagine that causes jerky animation. Or you have to be polling very frequently to observe FB changes. Weston will be using hardware planes. This means that it is not enough to sample the primary plane, you have to sample all planes and the full KMS state including the active areas of each FB. Then you get to composite those various FBs together before you have a single picture. You will probably also have problems with buffer content synchronization: you may be reading the content while Weston is already flipped to another FB and is half-way repainting the buffer you are reading. This will cause artifacts in the image. So I cannot really recommend that approach at all, based on a theoretical analysis. Thanks, pq
pgpp8kMrBW3nG.pgp
Description: OpenPGP digital signature
_______________________________________________ wayland-devel mailing list wayland-devel@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/wayland-devel
