[Mesa-dev] [PATCH] nvc0: bump the amount of shared memory per MP on Maxwell

[Ilia Mirkin]

Reviewed-by: Ilia Mirkin <imirkin at alum.mit.edu>

On Mon, Apr 25, 2016 at 6:17 PM, Samuel Pitoiset
<samuel.pitoiset at gmail.com> wrote:
> According to the CUDA compute capability version, GM10x can expose
> 64KB of shared memory while GM20x can use 96KB.
>
> Signed-off-by: Samuel Pitoiset <samuel.pitoiset at gmail.com>
> ---
>  src/gallium/drivers/nouveau/nvc0/nvc0_screen.c | 12 +++++++++++-
>  1 file changed, 11 insertions(+), 1 deletion(-)
>
> diff --git a/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c b/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c
> index ea670a6..999d52e 100644
> --- a/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c
> +++ b/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c
> @@ -444,7 +444,17 @@ nvc0_screen_get_compute_param(struct pipe_screen *pscreen,
>     case PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE: /* g[] */
>        RET((uint64_t []) { 1ULL << 40 });
>     case PIPE_COMPUTE_CAP_MAX_LOCAL_SIZE: /* s[] */
> -      RET((uint64_t []) { 48 << 10 });
> +      switch (obj_class) {
> +      case GM200_COMPUTE_CLASS:
> +         RET((uint64_t []) { 96 << 10 });
> +         break;
> +      case GM107_COMPUTE_CLASS:
> +         RET((uint64_t []) { 64 << 10 });
> +         break;
> +      default:
> +         RET((uint64_t []) { 48 << 10 });
> +         break;
> +      }
>     case PIPE_COMPUTE_CAP_MAX_PRIVATE_SIZE: /* l[] */
>        RET((uint64_t []) { 512 << 10 });
>     case PIPE_COMPUTE_CAP_MAX_INPUT_SIZE: /* c[], arbitrary limit */
> --
> 2.8.0
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev