[Mesa-dev] [PATCH] radeonsi/gfx9: add a temporary workaround for a tessellation driver bug

[Marek Olšák]

Ping.

On Wed, Aug 16, 2017 at 12:57 AM, Marek Olšák <maraeo at gmail.com> wrote:
> From: Marek Olšák <marek.olsak at amd.com>
>
> The workaround will do for now. The root cause is still unknown.
>
> This fixes new piglit: 16in-1out
>
> Cc: 17.1 17.2 <mesa-stable at lists.freedesktop.org>
> ---
>  src/gallium/drivers/radeonsi/si_state_draw.c | 6 +++++-
>  1 file changed, 5 insertions(+), 1 deletion(-)
>
> diff --git a/src/gallium/drivers/radeonsi/si_state_draw.c b/src/gallium/drivers/radeonsi/si_state_draw.c
> index f17f570..a26e38d 100644
> --- a/src/gallium/drivers/radeonsi/si_state_draw.c
> +++ b/src/gallium/drivers/radeonsi/si_state_draw.c
> @@ -187,21 +187,25 @@ static void si_emit_derived_tess_state(struct si_context *sctx,
>         /* Make sure the output data fits in the offchip buffer */
>         *num_patches = MIN2(*num_patches,
>                             (sctx->screen->tess_offchip_block_dw_size * 4) /
>                             output_patch_size);
>
>         /* Not necessary for correctness, but improves performance. The
>          * specific value is taken from the proprietary driver.
>          */
>         *num_patches = MIN2(*num_patches, 40);
>
> -       if (sctx->b.chip_class == SI) {
> +       if (sctx->b.chip_class == SI ||
> +           /* TODO: fix GFX9 where a threadgroup contains more than 1 wave and
> +            * LS vertices per patch > HS vertices per patch. Piglit: 16in-1out */
> +           (sctx->b.chip_class == GFX9 &&
> +            num_tcs_input_cp > num_tcs_output_cp)) {
>                 /* SI bug workaround, related to power management. Limit LS-HS
>                  * threadgroups to only one wave.
>                  */
>                 unsigned one_wave = 64 / MAX2(num_tcs_input_cp, num_tcs_output_cp);
>                 *num_patches = MIN2(*num_patches, one_wave);
>         }
>
>         /* The VGT HS block increments the patch ID unconditionally
>          * within a single threadgroup. This results in incorrect
>          * patch IDs when instanced draws are used.
> --
> 2.7.4
>