[Mesa-dev] [PATCH 1/2] radeonsi: fix a VGT hang with primitive restart on Polaris10 and later
Jakob Bornecrantz
jakob at collabora.com
Wed Oct 17 17:35:35 UTC 2018
Tested-by: Jakob Bornecrantz <jakob at collabora.com>
On Wed, Oct 17, 2018 at 5:29 PM Marek Olšák <maraeo at gmail.com> wrote:
>
> From: Marek Olšák <marek.olsak at amd.com>
>
> Cc: 18.1 18.2 <mesa-stable at lists.freedesktop.org>
> ---
> src/gallium/drivers/radeonsi/si_state_draw.c | 10 ++++++++--
> 1 file changed, 8 insertions(+), 2 deletions(-)
>
> diff --git a/src/gallium/drivers/radeonsi/si_state_draw.c b/src/gallium/drivers/radeonsi/si_state_draw.c
> index 83eb646b791..612ca910cb9 100644
> --- a/src/gallium/drivers/radeonsi/si_state_draw.c
> +++ b/src/gallium/drivers/radeonsi/si_state_draw.c
> @@ -376,21 +376,21 @@ si_get_init_multi_vgt_param(struct si_screen *sscreen,
> }
>
> if (sscreen->info.chip_class >= CIK) {
> /* WD_SWITCH_ON_EOP has no effect on GPUs with less than
> * 4 shader engines. Set 1 to pass the assertion below.
> * The other cases are hardware requirements.
> *
> * Polaris supports primitive restart with WD_SWITCH_ON_EOP=0
> * for points, line strips, and tri strips.
> */
> - if (sscreen->info.max_se < 4 ||
> + if (sscreen->info.max_se <= 2 ||
> key->u.prim == PIPE_PRIM_POLYGON ||
> key->u.prim == PIPE_PRIM_LINE_LOOP ||
> key->u.prim == PIPE_PRIM_TRIANGLE_FAN ||
> key->u.prim == PIPE_PRIM_TRIANGLE_STRIP_ADJACENCY ||
> (key->u.primitive_restart &&
> (sscreen->info.family < CHIP_POLARIS10 ||
> (key->u.prim != PIPE_PRIM_POINTS &&
> key->u.prim != PIPE_PRIM_LINE_STRIP &&
> key->u.prim != PIPE_PRIM_TRIANGLE_STRIP))) ||
> key->u.count_from_stream_output)
> @@ -407,35 +407,41 @@ si_get_init_multi_vgt_param(struct si_screen *sscreen,
> * instances are smaller than a primgroup.
> * Assume indirect draws always use small instances.
> * This is needed for good VS wave utilization.
> */
> if (sscreen->info.chip_class <= VI &&
> sscreen->info.max_se == 4 &&
> key->u.multi_instances_smaller_than_primgroup)
> wd_switch_on_eop = true;
>
> /* Required on CIK and later. */
> - if (sscreen->info.max_se > 2 && !wd_switch_on_eop)
> + if (sscreen->info.max_se == 4 && !wd_switch_on_eop)
> ia_switch_on_eoi = true;
>
> /* Required by Hawaii and, for some special cases, by VI. */
> if (ia_switch_on_eoi &&
> (sscreen->info.family == CHIP_HAWAII ||
> (sscreen->info.chip_class == VI &&
> (key->u.uses_gs || max_primgroup_in_wave != 2))))
> partial_vs_wave = true;
>
> /* Instancing bug on Bonaire. */
> if (sscreen->info.family == CHIP_BONAIRE && ia_switch_on_eoi &&
> key->u.uses_instancing)
> partial_vs_wave = true;
>
> + /* This only applies to Polaris10 and later 4 SE chips.
> + * wd_switch_on_eop is already true on all other chips.
> + */
> + if (!wd_switch_on_eop && key->u.primitive_restart)
> + partial_vs_wave = true;
> +
> /* If the WD switch is false, the IA switch must be false too. */
> assert(wd_switch_on_eop || !ia_switch_on_eop);
> }
>
> /* If SWITCH_ON_EOI is set, PARTIAL_ES_WAVE must be set too. */
> if (sscreen->info.chip_class <= VI && ia_switch_on_eoi)
> partial_es_wave = true;
>
> return S_028AA8_SWITCH_ON_EOP(ia_switch_on_eop) |
> S_028AA8_SWITCH_ON_EOI(ia_switch_on_eoi) |
> --
> 2.17.1
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
More information about the mesa-dev
mailing list