Nouveau - May 2020 - [RFC] Deprecate AGP GART support for Radeon/Nouveau/TTM

On 2020-05-20 4:43 p.m., Christian K?nig wrote:
> Am 13.05.20 um 13:03 schrieb Christian K?nig:
>> Unfortunately AGP is still to widely used as we could just drop
>> support for using its GART.
>>
>> Not using the AGP GART also doesn't mean a loss in functionality
since
>> drivers will just fallback to the driver specific PCI GART.
>>
>> For now just deprecate the code and don't enable the AGP GART in
TTM
>> even when general AGP support is available.
> 
> So I've used an ancient system (32bit) to setup a test box for this.
> 
> 
> The first GPU I could test is an RV280 (Radeon 9200 PRO) which is easily
> 15 years old.
> 
> What happens in AGP mode is that glxgears shows artifacts during
> rendering on this system.
> 
> In PCI mode those rendering artifacts are gone and glxgears seems to
> draw everything correctly now.
> 
> Performance is obviously not comparable, cause in AGP we don't render
> all triangles correctly.
> 
> 
> The second GPU I could test is an RV630 PRO (Radeon HD 2600 PRO AGP)
> which is more than 10 years old.
> 
> As far as I can tell this one works in both AGP and PCIe mode perfectly
> fine.
> 
> Since this is only a 32bit system I couldn't really test any OpenGL
game
> that well.
> 
> But for glxgears switching from AGP to PCIe mode seems to result in a
> roughly 5% performance drop.
> 
> The surprising reason for this is not the better TLB performance, but
> the lack of USWC support for the PCIe GART in radeon.
I suspect the main reason it's only 5% is that PCIe GART page tables are
stored in VRAM, so they don't need to be fetched across the PCIe link
(and presumably it has more than one TLB entry as well). The difference
is much bigger with native AGP ASICs with PCI GART.


-- 
Earthling Michel D?nzer               |               https://redhat.com
Libre software enthusiast             |             Mesa and X developer

Am 20.05.20 um 18:25 schrieb Michel D?nzer:
> On 2020-05-20 4:43 p.m., Christian K?nig wrote:
>> Am 13.05.20 um 13:03 schrieb Christian K?nig:
>>> Unfortunately AGP is still to widely used as we could just drop
>>> support for using its GART.
>>>
>>> Not using the AGP GART also doesn't mean a loss in
functionality since
>>> drivers will just fallback to the driver specific PCI GART.
>>>
>>> For now just deprecate the code and don't enable the AGP GART
in TTM
>>> even when general AGP support is available.
>> So I've used an ancient system (32bit) to setup a test box for
this.
>>
>>
>> The first GPU I could test is an RV280 (Radeon 9200 PRO) which is
easily
>> 15 years old.
>>
>> What happens in AGP mode is that glxgears shows artifacts during
>> rendering on this system.
>>
>> In PCI mode those rendering artifacts are gone and glxgears seems to
>> draw everything correctly now.
>>
>> Performance is obviously not comparable, cause in AGP we don't
render
>> all triangles correctly.
>>
>>
>> The second GPU I could test is an RV630 PRO (Radeon HD 2600 PRO AGP)
>> which is more than 10 years old.
>>
>> As far as I can tell this one works in both AGP and PCIe mode perfectly
>> fine.
>>
>> Since this is only a 32bit system I couldn't really test any OpenGL
game
>> that well.
>>
>> But for glxgears switching from AGP to PCIe mode seems to result in a
>> roughly 5% performance drop.
>>
>> The surprising reason for this is not the better TLB performance, but
>> the lack of USWC support for the PCIe GART in radeon.
> I suspect the main reason it's only 5% is that PCIe GART page tables
are
> stored in VRAM, so they don't need to be fetched across the PCIe link
> (and presumably it has more than one TLB entry as well). The difference
> is much bigger with native AGP ASICs with PCI GART.
Do you have some hardware you could give that a try on?

I mean the first card I put into the system obviously only worked 
correctly with AGP disabled.

While I agree that it means a performance regression, this is a rather 
high motivation to go ahead with at least the first patch.

Christian.

On 2020-05-22 12:40 p.m., Christian K?nig wrote:
> Am 20.05.20 um 18:25 schrieb Michel D?nzer:
>> On 2020-05-20 4:43 p.m., Christian K?nig wrote:
>>> Am 13.05.20 um 13:03 schrieb Christian K?nig:
>>>> Unfortunately AGP is still to widely used as we could just drop
>>>> support for using its GART.
>>>>
>>>> Not using the AGP GART also doesn't mean a loss in
functionality since
>>>> drivers will just fallback to the driver specific PCI GART.
>>>>
>>>> For now just deprecate the code and don't enable the AGP
GART in TTM
>>>> even when general AGP support is available.
>>> So I've used an ancient system (32bit) to setup a test box for
this.
>>>
>>>
>>> The first GPU I could test is an RV280 (Radeon 9200 PRO) which is
easily
>>> 15 years old.
>>>
>>> What happens in AGP mode is that glxgears shows artifacts during
>>> rendering on this system.
>>>
>>> In PCI mode those rendering artifacts are gone and glxgears seems
to
>>> draw everything correctly now.
>>>
>>> Performance is obviously not comparable, cause in AGP we don't
render
>>> all triangles correctly.
>>>
>>>
>>> The second GPU I could test is an RV630 PRO (Radeon HD 2600 PRO
AGP)
>>> which is more than 10 years old.
>>>
>>> As far as I can tell this one works in both AGP and PCIe mode
perfectly
>>> fine.
>>>
>>> Since this is only a 32bit system I couldn't really test any
OpenGL game
>>> that well.
>>>
>>> But for glxgears switching from AGP to PCIe mode seems to result in
a
>>> roughly 5% performance drop.
>>>
>>> The surprising reason for this is not the better TLB performance,
but
>>> the lack of USWC support for the PCIe GART in radeon.
>> I suspect the main reason it's only 5% is that PCIe GART page
tables are
>> stored in VRAM, so they don't need to be fetched across the PCIe
link
>> (and presumably it has more than one TLB entry as well). The difference
>> is much bigger with native AGP ASICs with PCI GART.
> 
> Do you have some hardware you could give that a try on?
As I mentioned before, I tested this many times on my AGP PowerBooks
back in the day. The result was always a similar, big hit with PCI GART
vs AGP (even just 1x). I haven't seen any reason to believe this has
changed.

> While I agree that it means a performance regression, this is a rather
> high motivation to go ahead with at least the first patch.
I totally agree with the benefits, I just want everyone to be honest and
clear about the performance hit with native AGP Radeons, which already
have very weak performance by today's standards even with AGP.


-- 
Earthling Michel D?nzer               |               https://redhat.com
Libre software enthusiast             |             Mesa and X developer