Re: [Mesa-dev] [PATCH 0/7] broadcom/vc4: Native fence fd support

[Eric Anholt]

Stefan Schake  writes:

> This series adds support for the native fence fd extension to vc4.
> The implementation relies on a newly introduced kernel interface that
> allows submitting syncobjs for in/out fences during job submission.
>
> Since syncobjs are relatively new, patches 1 and 3 have build changes
> for automake and meson to require a recent libdrm version.
>
> There is some scope here to replace the previous sequence number based
> implementation with only syncobjs, but given that we need to continue
> supporting older kernels I felt it would have only added complexity.
>
> This has been tested with piglit and kmscube -A(tomic). In particular,
> I checked that the fd numbers weren't increasing during the kmscube run.
>
> Should not be merged until the kernel side lands. In particular, patch 2
> is purely provisional.

A few comments here that I think will clean up the series a bit, but it
looks pretty good to me.  In particular, I think the kernel ABI will be
fine, with the little comments I had in the kernel series.

Thanks for working on this!


signature.asc
Description: PGP signature
___
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev

Re: [Mesa-dev] [PATCH 0/7] broadcom/vc4: Native fence fd support

[Stefan Schake]

On Sun, Apr 22, 2018 at 12:41 AM, Stefan Schake  wrote:
> This series adds support for the native fence fd extension to vc4.
> The implementation relies on a newly introduced kernel interface that
> allows submitting syncobjs for in/out fences during job submission.
>
> Since syncobjs are relatively new, patches 1 and 3 have build changes
> for automake and meson to require a recent libdrm version.
>
> There is some scope here to replace the previous sequence number based
> implementation with only syncobjs, but given that we need to continue
> supporting older kernels I felt it would have only added complexity.
>
> This has been tested with piglit and kmscube -A(tomic). In particular,
> I checked that the fd numbers weren't increasing during the kmscube run.
>
> Should not be merged until the kernel side lands. In particular, patch 2
> is purely provisional.
>
> Stefan Schake (7):
>   broadcom/vc4: Drop libdrm_vc4 requirement
>   drm-uapi: Update vc4 header with syncobj submit support
>   broadcom/vc4: Bump libdrm requirement
>   broadcom/vc4: Detect syncobj support
>   broadcom/vc4: Store job fence in syncobj
>   broadcom/vc4: Expose syncobj import
>   broadcom/vc4: Native fence fd support
>
>  configure.ac|  3 +-
>  include/drm-uapi/vc4_drm.h  | 15 
>  meson.build |  2 +
>  src/gallium/drivers/vc4/vc4_context.c   | 51 ++---
>  src/gallium/drivers/vc4/vc4_context.h   | 12 +-
>  src/gallium/drivers/vc4/vc4_fence.c | 68 
> +++--
>  src/gallium/drivers/vc4/vc4_job.c   | 50 +++-
>  src/gallium/drivers/vc4/vc4_screen.c|  8 +++-
>  src/gallium/drivers/vc4/vc4_screen.h|  5 ++-
>  src/gallium/drivers/vc4/vc4_simulator.c |  1 +
>  10 files changed, 198 insertions(+), 17 deletions(-)
>
> --
> 2.14.1
>

Kernel series is here:

https://patchwork.freedesktop.org/series/42082/
___
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev

[Mesa-dev] [PATCH 0/7] broadcom/vc4: Native fence fd support

[Stefan Schake]

This series adds support for the native fence fd extension to vc4.
The implementation relies on a newly introduced kernel interface that
allows submitting syncobjs for in/out fences during job submission.

Since syncobjs are relatively new, patches 1 and 3 have build changes
for automake and meson to require a recent libdrm version.

There is some scope here to replace the previous sequence number based
implementation with only syncobjs, but given that we need to continue
supporting older kernels I felt it would have only added complexity.

This has been tested with piglit and kmscube -A(tomic). In particular,
I checked that the fd numbers weren't increasing during the kmscube run.

Should not be merged until the kernel side lands. In particular, patch 2
is purely provisional.

Stefan Schake (7):
  broadcom/vc4: Drop libdrm_vc4 requirement
  drm-uapi: Update vc4 header with syncobj submit support
  broadcom/vc4: Bump libdrm requirement
  broadcom/vc4: Detect syncobj support
  broadcom/vc4: Store job fence in syncobj
  broadcom/vc4: Expose syncobj import
  broadcom/vc4: Native fence fd support

 configure.ac|  3 +-
 include/drm-uapi/vc4_drm.h  | 15 
 meson.build |  2 +
 src/gallium/drivers/vc4/vc4_context.c   | 51 ++---
 src/gallium/drivers/vc4/vc4_context.h   | 12 +-
 src/gallium/drivers/vc4/vc4_fence.c | 68 +++--
 src/gallium/drivers/vc4/vc4_job.c   | 50 +++-
 src/gallium/drivers/vc4/vc4_screen.c|  8 +++-
 src/gallium/drivers/vc4/vc4_screen.h|  5 ++-
 src/gallium/drivers/vc4/vc4_simulator.c |  1 +
 10 files changed, 198 insertions(+), 17 deletions(-)

-- 
2.14.1

___
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev