AMDGPU+RadeonSI With Deus Ex: Mankind Divided, One Week Later

[mibo]

So, one week and I still can't get the game even starting.
I also would like to test it with different MESA versions, but I already fail to give it the libraries it wants to have :-(

I saw several posts in the Steam discussion forum from Arch and Mint users who also could not get it starting (links below).
I'm on OpenSUSE 42.1 and somehow it has to work because Pontostroy did quickly post game benchmarks on his web site ( http://www.gearsongallium.com/?p=3467 ).

My problem seems to be that several libraries are found but seem to be somehow wrong versions?

./DeusExMD: /lib64/libssl.so.1.0.0: no version information available (required by ./DeusExMD)
./DeusExMD: /lib64/libcrypto.so.1.0.0: no version information available (required by ./DeusExMD)
./DeusExMD: /usr/lib64/libldap_r-2.4.so.2: no version information available (required by /home/michael/.local/share/Steam/steamapps/common/Deus Ex Mankind Divided/bin/../lib/x86_64/libcurl.so.4)
./DeusExMD: /usr/lib64/liblber-2.4.so.2: no version information available (required by /home/michael/.local/share/Steam/steamapps/common/Deus Ex Mankind Divided/bin/../lib/x86_64/libcurl.so.4)

Could somebody please help a noob to get it running?

Thanks.

https://steamcommunity.com/app/33700...7094327954264/
https://steamcommunity.com/app/33700...7094313537819/

[faldzip]

Seems not that bad with AMDGPU-PRO blob (the previous version) with R7 260X:

http://openbenchmarking.org/result/1611035-TA-DEUSEXMAN52,1611134-HA-DXMD860KR18

compared to Nvidia cards like 760 (I guess that it could even beat the 750ti). But it is hard to name those results as 'playable'...

[pal666]

Actually most awaited feature would be inclusion of S3TC, since patent expire next year... less than 11 month to go

s3tc is included in separate package, no need to await it

[pal666]

There no bugs that affects gaming in Oibaf ppa llvm 3.9. Prove it if there is.

you should prove your claims instead of asking someone to disprove them

[pal666]

please help

you shold ask someone with your distro

[mibo]

you shold ask someone with your distro

Good point.
As I showed with the links, also users of other distros seem to be affected. Therefore, I try to find some help here.

[dungeon]

s3tc is included in separate package, no need to await it

Of course you can, but how many actually do it Some people even file bugs, without realising they don't have this installed

Anything which leads to unneeded diversify is not good, so we can only celebrate when this get integrated into mesa

[faldzip]

Seems not that bad with AMDGPU-PRO blob (the previous version) with R7 260X:

http://openbenchmarking.org/result/1611035-TA-DEUSEXMAN52,1611134-HA-DXMD860KR18

compared to Nvidia cards like 760 (I guess that it could even beat the 750ti). But it is hard to name those results as 'playable'...

For comparison I did some tests on Windows 10 on the same rig and I get (in DX12, in DX11 it is 1-2 fps worse):
Low: 37.3fps
High: 24.1fps

So Win10+High == Ubuntu+Low...

[marek]

Then why experts say that Bindless alone is half of the AZDO gain?

What experts? Marek doesn't say that.

It can't half for everybody. It depends on the driver architecture and the kernel driver interfaces.

It also can't be half, because MultiDrawIndirect alone can give you 10x gain easily. For Mesa, bindless can give +5-20% maybe.

[artivision]

What experts? Marek doesn't say that.

It can't half for everybody. It depends on the driver architecture and the kernel driver interfaces.

It also can't be half, because MultiDrawIndirect alone can give you 10x gain easily. For Mesa, bindless can give +5-20% maybe.

MultiDrawIndirect and other AZDO futures cannot work to their best without some fancy low level futures. Bindless is a must and not only 20% less overhead. Start from half (and lower) if you want to be accurate. Except if your ASIC cannot do it right, then just say so.