[Mesa-dev] [PATCH 4/4] anv/blorp: Flush the texture cache in UpdateBuffer

Lionel Landwerlin lionel.g.landwerlin at intel.com
Wed Apr 12 21:30:25 UTC 2017


On 12/04/17 12:57, Jason Ekstrand wrote:
> On Wed, Apr 12, 2017 at 12:25 PM, Lionel Landwerlin 
> <lionel.g.landwerlin at intel.com <mailto:lionel.g.landwerlin at intel.com>> 
> wrote:
>
>     On 31/03/17 16:17, Jason Ekstrand wrote:
>
>         Cc: "13.0 17.0" <mesa-stable at lists.freedesktop.org
>         <mailto:mesa-stable at lists.freedesktop.org>>
>         ---
>           src/intel/vulkan/anv_blorp.c | 5 +++++
>           1 file changed, 5 insertions(+)
>
>         diff --git a/src/intel/vulkan/anv_blorp.c
>         b/src/intel/vulkan/anv_blorp.c
>         index 72a468a..f26f5e5 100644
>         --- a/src/intel/vulkan/anv_blorp.c
>         +++ b/src/intel/vulkan/anv_blorp.c
>         @@ -690,6 +690,11 @@ void anv_CmdUpdateBuffer(
>                assert(max_update_size < MAX_SURFACE_DIM * 4);
>           +   /* We're about to read data that was written from the
>         CPU.  Flush the
>
>     Did you mean GPU?
>
>
> No, I mean CPU.  We copy the data provided by the user into a chunk of 
> the dynamic state pool and then use blorp to copy it to the buffer 
> specified by the user.

Right, for some reason the phrasing got me confused :/

>         +    * texture cache so we don't get anything stale.
>         +    */
>         +   cmd_buffer->state.pending_pipe_bits |=
>         ANV_PIPE_TEXTURE_CACHE_INVALIDATE_BIT;
>
>
>     Why just the texture cache? Should we add data cache + render target?
>
>
> Because we're not writing from the GPU and we're not using image load 
> store.

Thanks, I should have read the do_buffer_copy()...

I think we might need a anv_state_flush() on tmp_data.

With that :

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin at intel.com>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/mesa-dev/attachments/20170412/5e361067/attachment.html>


More information about the mesa-dev mailing list