[Mesa-dev] [PATCH] radv: use simpler indirect packet 3 if possible.
Nicolai Hähnle
nhaehnle at gmail.com
Mon Sep 11 13:48:45 UTC 2017
On 07.09.2017 15:07, Bas Nieuwenhuizen wrote:
>
>
> On Thu, Sep 7, 2017, at 14:52, Nicolai Hähnle wrote:
>> On 07.09.2017 12:26, Dave Airlie wrote:
>>>
>>>
>>> On 7 Sep. 2017 6:34 pm, "Nicolai Hähnle" <nhaehnle at gmail.com
>>> <mailto:nhaehnle at gmail.com>> wrote:
>>>
>>> On 07.09.2017 09 <tel:07.09.2017%2009>:58, Bas Nieuwenhuizen wrote:
>>>
>>> I'm not really happy with this, what happens if a game actually
>>> uses
>>> e.g. indirect count in a secondary cmd buf?
>>>
>>>
>>> Note that some packets that require fetching data in the CP cannot
>>> be run in secondary command buffers, because the data fetching path
>>> overlaps with that used for fetching IB2. I thought this only
>>> affects SET_PREDICATION and OCCLUSION_QUERY though.
>>>
>>>
>>>
>>> Can you check it, it may be multi indirect is busted on some firmwares,
>>> this seems cik specific.
>>
>> Is this IB1 or IB2? I just checked, and at least the fetcher problem
>> affecting SET_PREDICATION should not affect the various multi draw
>> packets.
>>
>
> I'm not familiar with the IB1 vs. IB2 terminology. It is a nested IB
> using PKT3_INDIRECT_BUFFER_CIK without chain. (i.e. the IB submitted in
> the IOCTL calls another IB with the PKT3_INDIRECT_BUFFER_CIK, which does
> the multi draw).
Right, that's IB2. And apparently it doesn't support indirect draws
after all, sorry.
Cheers,
Nicolai
>>
>>> Also is set predication a problem for all Pred setting types?
>>
>> Huh, it looks like the DX12/Vulkan/Bool mode actually takes a different
>> path. So only occlusion queries and streamout overflow are affected. TIL.
>>
>> Cheers,
>> Nicolai
>>
>>
>>>
>>> Dave.
>>>
>>>
>>> That said this patch seems correct, so
>>>
>>> Reviewed-by: Bas Nieuwenhuizen <bas at basnieuwenhuizen.nl
>>> <mailto:bas at basnieuwenhuizen.nl>>
>>>
>>> You may want to nominate this for stable (and provide a backport
>>> probably..)
>>>
>>> On Thu, Sep 7, 2017, at 05:03, Dave Airlie wrote:
>>>
>>> From: Dave Airlie <airlied at redhat.com
>>> <mailto:airlied at redhat.com>>
>>>
>>> This fixes some observed hangs on CIK GPUs.
>>>
>>> Signed-off-by: Dave Airlie <airlied at redhat.com
>>> <mailto:airlied at redhat.com>>
>>> ---
>>> src/amd/vulkan/radv_cmd_buffer.c | 37
>>> +++++++++++++++++++++++--------------
>>> 1 file changed, 23 insertions(+), 14 deletions(-)
>>>
>>> diff --git a/src/amd/vulkan/radv_cmd_buffer.c
>>> b/src/amd/vulkan/radv_cmd_buffer.c
>>> index b372123..bc4aeb3 100644
>>> --- a/src/amd/vulkan/radv_cmd_buffer.c
>>> +++ b/src/amd/vulkan/radv_cmd_buffer.c
>>> @@ -2834,20 +2834,29 @@ radv_cs_emit_indirect_draw_packet(struct
>>> radv_cmd_buffer *cmd_buffer,
>>> uint32_t base_reg =
>>> cmd_buffer->state.pipeline->graphics.vtx_base_sgpr;
>>> assert(base_reg);
>>> - radeon_emit(cs, PKT3(indexed ?
>>> PKT3_DRAW_INDEX_INDIRECT_MULTI :
>>> - PKT3_DRAW_INDIRECT_MULTI,
>>> - 8, false));
>>> - radeon_emit(cs, 0);
>>> - radeon_emit(cs, (base_reg - SI_SH_REG_OFFSET) >> 2);
>>> - radeon_emit(cs, ((base_reg + 4) - SI_SH_REG_OFFSET)
>>> >> 2);
>>> - radeon_emit(cs, (((base_reg + 8) - SI_SH_REG_OFFSET)
>>> >> 2) |
>>> -
>>> S_2C3_DRAW_INDEX_ENABLE(draw_id_enable) |
>>> -
>>> S_2C3_COUNT_INDIRECT_ENABLE(!!count_va));
>>> - radeon_emit(cs, draw_count); /* count */
>>> - radeon_emit(cs, count_va); /* count_addr */
>>> - radeon_emit(cs, count_va >> 32);
>>> - radeon_emit(cs, stride); /* stride */
>>> - radeon_emit(cs, di_src_sel);
>>> + if (draw_count == 1 && !count_va && !draw_id_enable) {
>>> + radeon_emit(cs, PKT3(indexed ?
>>> PKT3_DRAW_INDEX_INDIRECT :
>>> + PKT3_DRAW_INDIRECT, 3,
>>> false));
>>> + radeon_emit(cs, 0);
>>> + radeon_emit(cs, (base_reg -
>>> SI_SH_REG_OFFSET) >> 2);
>>> + radeon_emit(cs, ((base_reg + 4) -
>>> SI_SH_REG_OFFSET) >>
>>> 2);
>>> + radeon_emit(cs, di_src_sel);
>>> + } else {
>>> + radeon_emit(cs, PKT3(indexed ?
>>> PKT3_DRAW_INDEX_INDIRECT_MULTI :
>>> + PKT3_DRAW_INDIRECT_MULTI,
>>> + 8, false));
>>> + radeon_emit(cs, 0);
>>> + radeon_emit(cs, (base_reg -
>>> SI_SH_REG_OFFSET) >> 2);
>>> + radeon_emit(cs, ((base_reg + 4) -
>>> SI_SH_REG_OFFSET) >>
>>> 2);
>>> + radeon_emit(cs, (((base_reg + 8) -
>>> SI_SH_REG_OFFSET) >>
>>> 2) |
>>> +
>>> S_2C3_DRAW_INDEX_ENABLE(draw_id_enable) |
>>> +
>>> S_2C3_COUNT_INDIRECT_ENABLE(!!count_va));
>>> + radeon_emit(cs, draw_count); /* count */
>>> + radeon_emit(cs, count_va); /* count_addr */
>>> + radeon_emit(cs, count_va >> 32);
>>> + radeon_emit(cs, stride); /* stride */
>>> + radeon_emit(cs, di_src_sel);
>>> + }
>>> }
>>> static void
>>> --
>>> 2.9.4
>>>
>>> _______________________________________________
>>> mesa-dev mailing list
>>> mesa-dev at lists.freedesktop.org
>>> <mailto:mesa-dev at lists.freedesktop.org>
>>> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
>>> <https://lists.freedesktop.org/mailman/listinfo/mesa-dev>
>>>
>>> _______________________________________________
>>> mesa-dev mailing list
>>> mesa-dev at lists.freedesktop.org
>>> <mailto:mesa-dev at lists.freedesktop.org>
>>> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
>>> <https://lists.freedesktop.org/mailman/listinfo/mesa-dev>
>>>
>>>
>>>
>>> --
>>> Lerne, wie die Welt wirklich ist,
>>> Aber vergiss niemals, wie sie sein sollte.
>>>
>>> _______________________________________________
>>> mesa-dev mailing list
>>> mesa-dev at lists.freedesktop.org <mailto:mesa-dev at lists.freedesktop.org>
>>> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
>>> <https://lists.freedesktop.org/mailman/listinfo/mesa-dev>
>>>
>>>
>>
>>
>> --
>> Lerne, wie die Welt wirklich ist,
>> Aber vergiss niemals, wie sie sein sollte.
--
Lerne, wie die Welt wirklich ist,
Aber vergiss niemals, wie sie sein sollte.
More information about the mesa-dev
mailing list