[igt-dev] [PATCH i-g-t 1/3] lib/i915_blt: Remove src == dst pitch restriction

[Karolina Stolarek]

On 15.12.2022 09:41, Karolina Stolarek wrote:
> On 14.12.2022 19:57, Zbigniew Kempczyński wrote:
>> On Tue, Dec 13, 2022 at 04:37:26PM +0100, Karolina Stolarek wrote:
>>> On 12.12.2022 13:50, Zbigniew Kempczyński wrote:
>>>> During debugging phase we established there's not necessary to enforce
>>>> same pitch for destination surface in FULL_RESOLVE mode. Relax this
>>>> condition allowing caller to pass requirement surface configuration.
>>>>
>>>> Signed-off-by: Zbigniew Kempczyński <zbigniew.kempczynski at intel.com>
>>>> ---
>>>>    lib/i915/i915_blt.c | 8 +++-----
>>>>    1 file changed, 3 insertions(+), 5 deletions(-)
>>>>
>>>> diff --git a/lib/i915/i915_blt.c b/lib/i915/i915_blt.c
>>>> index 3776c56c60..42c28623f9 100644
>>>> --- a/lib/i915/i915_blt.c
>>>> +++ b/lib/i915/i915_blt.c
>>>> @@ -317,13 +317,11 @@ static void fill_data(struct 
>>>> gen12_block_copy_data *data,
>>>>        data->dw00.special_mode = __special_mode(blt);
>>>>        data->dw00.length = extended_command ? 20 : 10;
>>>> -    if (__special_mode(blt) == SM_FULL_RESOLVE && blt->src.tiling 
>>>> == T_TILE64) {
>>>> -        data->dw01.dst_pitch = blt->src.pitch - 1;
>>>> +    if (__special_mode(blt) == SM_FULL_RESOLVE)
>>>
>>> "blt->src.tiling == T_TILE64" part was introduced in commit cdc0258a4672
>>> ("lib/i915_blt: Narrow setting same pitch and aux-ccs to Tile64 only").
>>> I know it's a bit nit-picky, but could we revert that change (as it 
>>> looks
>>> like it's not necessary), and have a separate commit that removes
>>> data->dw01.dst_pitch = blt->src.pitch - 1?
>>
>> Two reverts would be necessary: cdc0258a467 and 229bb0accbb7 so single 
>> commit
>> against three looks better for me. Tiling inplace testing is still in 
>> progress.
>> If you think better is to revert above and add another commit I'll do it.
> 
> To me, some of the 229bb0accbb7 is still here (setting aux_mode), so I'd 
> say the revert of it is not necessary. But, yeah, I think it would be 
> better to revert cdc0258a467.

After further discussions with Zbigniew, we decided not to revert 
cdc0258a467. The original implementation isn't correct, and reverting 
two patches with a change on the top of it only muddies the waters. 
Let's keep things simple, and go with this patch.

All the best,
Karolina

>>
>>>
>>> The question remains -- what actually caused the regression for Tile4 
>>> and
>>> xmajor? Was it because of the pitch settings? I run the tests a 
>>> couple of
>>> times in the row after applying this patch and everything worked as
>>> expected.
>>
>> Yes, dst pitch influences on destination surface state. For cascaded
>> src->mid->dst mid.pitch may differ from dst.pitch (tileX -> linear blit).
>> For inplace there's some side effect which we're debugging now.
> 
> OK, so that seems like a root cause. As for inplace, I think it can be 
> slightly different, but we can update the function once debugging is 
> finished.
> 
> All the best,
> Karolina
> 
>>
>> -- 
>> Zbigniew
>>
>>>
>>> All the best,
>>> Karolina
>>>
>>>>            data->dw01.dst_aux_mode = __aux_mode(&blt->src);
>>>> -    } else {
>>>> -        data->dw01.dst_pitch = blt->dst.pitch - 1;
>>>> +    else
>>>>            data->dw01.dst_aux_mode = __aux_mode(&blt->dst);
>>>> -    }
>>>> +    data->dw01.dst_pitch = blt->dst.pitch - 1;
>>>>        data->dw01.dst_mocs = blt->dst.mocs;
>>>>        data->dw01.dst_compression = blt->dst.compression;