[Mesa-dev] [PATCH] gallium/u_threaded: increase batch size to increase performance

[Kenneth Graunke]

On Saturday, September 1, 2018 12:14:38 AM PDT Marek Olšák wrote:
> From: Marek Olšák <marek.olsak at amd.com>
> 
> This reduces mutex overhead.
> 
> +4.4% performance with piglit/drawoverhead, DrawElements, Ryzen X1700

+14% with piglit/drawoverhead and iris_dri.so, DrawArrays, i7 7700HQ.

Acked-by: Kenneth Graunke <kenneth at whitecape.org>

> ---
>  src/gallium/auxiliary/util/u_threaded_context.h | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/src/gallium/auxiliary/util/u_threaded_context.h b/src/gallium/auxiliary/util/u_threaded_context.h
> index 53c5a7e8c4c..be6933d05a4 100644
> --- a/src/gallium/auxiliary/util/u_threaded_context.h
> +++ b/src/gallium/auxiliary/util/u_threaded_context.h
> @@ -213,21 +213,21 @@ struct tc_unflushed_batch_token;
>   * slots.
>   */
>  #define TC_MAX_BATCHES        10
>  
>  /* The size of one batch. Non-trivial calls (i.e. not setting a CSO pointer)
>   * can occupy multiple call slots.
>   *
>   * The idea is to have batches as small as possible but large enough so that
>   * the queuing and mutex overhead is negligible.
>   */
> -#define TC_CALLS_PER_BATCH    192
> +#define TC_CALLS_PER_BATCH    768
>  
>  /* Threshold for when to use the queue or sync. */
>  #define TC_MAX_STRING_MARKER_BYTES  512
>  
>  /* Threshold for when to enqueue buffer/texture_subdata as-is.
>   * If the upload size is greater than this, it will do instead:
>   * - for buffers: DISCARD_RANGE is done by the threaded context
>   * - for textures: sync and call the driver directly
>   */
>  #define TC_MAX_SUBDATA_BYTES        320
> 

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: This is a digitally signed message part.
URL: <https://lists.freedesktop.org/archives/mesa-dev/attachments/20180901/3ecd203f/attachment.sig>