Discussion: the ati/radeon open source drivers & radeon repository

Cdh · 2013-05-23 09:32:17

Does anyone know why htis is?

 ~/opencl-example (git)-[master] % ./hello_world
radeon: Failed to get PCI ID, error number -13
There are 1 platforms.
clGetDeviceIDs() failed: CL_DEVICE_NOT_FOUND
1 ~/opencl-example (git)-[master] % sudo ./hello_world
There are 1 platforms.
There are 1 GPU devices.
clCreateContext() succeeded.
clCreateCommandQueue() succeeded.
clCreateProgramWithSource() suceeded.
clBuildProgram() suceeded.
clCreateKernel() suceeded.
clCreateBuffer() succeeded.
clSetKernelArg() succeeded.
clEnqueueNDRangeKernel() suceeded.
clFinish() succeeded.
clEnqueueReadBuffer() suceeded.
pi = 3.141590
 ~/opencl-example (git)-[master] % ls -l /dev/dri/card*
crw-rw----+ 1 root video 226, 0 23. Mai 09:56 /dev/dri/card0
crw-rw----+ 1 root video 226, 1 21. Mai 14:26 /dev/dri/card1
 ~/opencl-example (git)-[master] % groups | grep -o video
video
 ~/opencl-example (git)-[master] %

The troubleshooting only says this is when you don't have access to /dev/dri/cardX but I am in the video group and have rw.
http://dri.freedesktop.org/wiki/GalliumCompute/

radeonsi:

6 passes, 65 fails

Not very good.

Last edited by Cdh (2013-05-23 09:35:01)

lordheavy · 2013-05-23 10:57:37

Cdh wrote:

Does anyone know why htis is?

 ~/opencl-example (git)-[master] % ./hello_world
radeon: Failed to get PCI ID, error number -13
There are 1 platforms.
clGetDeviceIDs() failed: CL_DEVICE_NOT_FOUND

Perhaps running it with ltrace/strace could give you some tips....

Cdh · 2013-05-23 11:29:47

without root: http://pastebin.com/raw.php?i=48RVLb1v
with root: http://pastebin.com/hxm0usPR

The difference seems to be at

ioctl(4, 0xc0406400, 0x11e7e60)         = 0
ioctl(4, 0xc0406400, 0x11e7e60)         = 0
ioctl(4, 0xc0106467, 0x7fff078f2ed0)    = -1 EACCES (Permission denied)

So, some special file/device access on some memory location? Not really sure how that works.

splashy · 2013-05-23 16:19:28

Is it ok if I pacman -Rdd libcl now that ati-dri-git provide /usr/lib/libOpenCL.so /usr/lib/libOpenCL.so.1 and /usr/lib/libOpenCL.so.1.0. ?

lordheavy · 2013-05-23 16:34:50

splashy wrote:

Is it ok if I pacman -Rdd libcl now that ati-dri-git provide /usr/lib/libOpenCL.so /usr/lib/libOpenCL.so.1 and /usr/lib/libOpenCL.so.1.0. ?

No problems to replace it

lordheavy · 2013-05-24 17:20:16

splashy wrote:

Is it ok if I pacman -Rdd libcl now that ati-dri-git provide /usr/lib/libOpenCL.so /usr/lib/libOpenCL.so.1 and /usr/lib/libOpenCL.so.1.0. ?

With today update, OpenCL support was completly remove from mesa-git package and moved to opencl-mesa-git package.
Now you don't have anymore to remove libcl and opencl-headers if you don't want use mesa's OpenCL solution.

New package: opencl-mesa-git

agapito · 2013-06-26 17:38:39

AMD Has Massive Radeon Patch Set - Power Management!

http://www.phoronix.com/scan.php?page=n … px=MTM5NjE

Great news!!

Cdh · 2013-06-26 19:29:02

To try it:

Use linux-git from AUR and replace git URL with 'git://people.freedesktop.org/~agd5f/linux#branch=drm-next-3.11-wip'

Download http://people.freedesktop.org/~agd5f/ra … RN_smc.bin and put it in /usr/lib/firmware/radeon
edit: Or the equivalent *_smc.bin for your card. PITCAIRN is the 7970M!

Last edited by Cdh (2013-06-29 06:33:05)

lordheavy · 2013-06-26 19:39:06

You can find the kernel linux-agd5f in mesa-git repo.
To enable kms you need also to install new firmware from http://people.freedesktop.org/~agd5f/radeon_ucode/ and install it in /usr/lib/firmware/radeon
Also add radeon.dpm=1 in the kernel command line.

from my syslinux.cfg file:

LABEL linuxagd5f
        MENU LABEL Linux agd5f kernel
        LINUX ../vmlinuz-linux-agd5f
        APPEND root=UUID=c7811590-ddcb-4f0e-9e25-b0f1185ae525 ro radeon.dpm=1
        INITRD ../initramfs-linux-agd5f.img

Have fun!

PS: it's a temporary solution while waiting linux-mainline-3.11rc1

Last edited by lordheavy (2013-06-26 19:39:54)

agapito · 2013-06-26 19:42:19

Thanks!!!! I will test it.

agapito · 2013-06-27 01:46:19

http://cgit.freedesktop.org/~agd5f/linu … 3.11-wip-2

More work in progress.

Last edited by agapito (2013-06-27 01:50:27)

lordheavy · 2013-06-27 06:25:07

agapito wrote:

http://cgit.freedesktop.org/~agd5f/linu … 3.11-wip-2
More work in progress.

Update to drm-next-3.11-wip-3 in progress

Ledti · 2013-06-27 09:46:32

lordheavy wrote:

agapito wrote:
http://cgit.freedesktop.org/~agd5f/linu … 3.11-wip-2
More work in progress.
Update to drm-next-3.11-wip-3 in progress

Wow, this is great. I didn't know there was a mesa-git repository. Thanks for providing this.

Anyway, for those curious to the benefits of the power management patches... wow.

Before the changes, my HD6870 card was idling at 66°C. After rebooting that went down to 45°C, and after ten more minutes it went down to 43°C. I also tried out Unreal's webgl demo and noticed that the temperature only rose a few degrees while it was playing. I don't know how graphics intensive that is (if at all, I'm kind of suspicious that it's mostly cpu) but I'm impressed. This is a huge step in the right direction and these changes seem to be on par or possibly even better than catalyst's power managment.

I also had to install xf86-video-ati-git from lordheavy's repo to use the new feature, otherwise my screen would just show artifacts after booting.

Last edited by Ledti (2013-06-27 11:43:31)

agapito · 2013-06-27 17:08:44

lordheavy wrote:

agapito wrote:
http://cgit.freedesktop.org/~agd5f/linu … 3.11-wip-2
More work in progress.
Update to drm-next-3.11-wip-3 in progress

More work in progress

http://cgit.freedesktop.org/~agd5f/linu … 3.11-wip-4

Akuma · 2013-06-27 20:27:46

lordheavy wrote:

You can find the kernel linux-agd5f in mesa-git repo.
To enable kms you need also to install new firmware from http://people.freedesktop.org/~agd5f/radeon_ucode/ and install it in /usr/lib/firmware/radeon
Also add radeon.dpm=1 in the kernel command line.

Thanks for the repo !

I just tried this on my laptop (RV635/HD3650), hoping it would stop being a frying pan. It works fine without radeon.dpm=1 (Composite/3D/DRI/KMS...) however adding the parameter makes it hang at boot with a black screen (precisely when login prompt should appear) even without starting Xorg. Yes I also installed the xf86-video-ati from mesa-git.
I wonder, isn't libdrm-git needed too ?

Update : same issue using libdrm-git

Last edited by Akuma (2013-06-27 21:57:01)

Stunts · 2013-06-27 21:35:47

@Akuma:
I have the same card as you and I will be testing this soon. I hope I will be able to provide some feedback soon.

agapito · 2013-06-28 21:06:56

Akuma wrote:

lordheavy wrote:
You can find the kernel linux-agd5f in mesa-git repo.
To enable kms you need also to install new firmware from http://people.freedesktop.org/~agd5f/radeon_ucode/ and install it in /usr/lib/firmware/radeon
Also add radeon.dpm=1 in the kernel command line.
Thanks for the repo !
I just tried this on my laptop (RV635/HD3650), hoping it would stop being a frying pan. It works fine without radeon.dpm=1 (Composite/3D/DRI/KMS...) however adding the parameter makes it hang at boot with a black screen (precisely when login prompt should appear) even without starting Xorg. Yes I also installed the xf86-video-ati from mesa-git.
I wonder, isn't libdrm-git needed too ?
Update : same issue using libdrm-git

I had the same problem with kernel wip1, but now is working with the latest kernels (> wip3)

Stunts · 2013-06-28 22:49:35

Cdh wrote:

To try it:
Use linux-git from AUR and replace git URL with 'git://people.freedesktop.org/~agd5f/linux#branch=drm-next-3.11-wip'
Download http://people.freedesktop.org/~agd5f/ra … RN_smc.bin and put it in /usr/lib/firmware/radeon

Crap! Using this, with wip5, all goes to hell!
If I do not set the option "radeon.dpm=1", everything works just fine. And by works fine, I mean the machine boots, but I get no power management on the card.
However, if I set the mentioned option, when modeseting was supposed to happen, and all I get is a "burn in" screen (yellow fading into white) and the laptop fully crashes.
Any ideas for this one? Or do you think it's worth a bug report?

Last edited by Stunts (2013-06-28 22:52:11)

Ledti · 2013-06-28 23:46:43

Stunts wrote:

Cdh wrote:
To try it:
Use linux-git from AUR and replace git URL with 'git://people.freedesktop.org/~agd5f/linux#branch=drm-next-3.11-wip'
Download http://people.freedesktop.org/~agd5f/ra … RN_smc.bin and put it in /usr/lib/firmware/radeon
Crap! Using this, with wip5, all goes to hell!
If I do not set the option "radeon.dpm=1", everything works just fine. And by works fine, I mean the machine boots, but I get no power management on the card.
However, if I set the mentioned option, when modeseting was supposed to happen, and all I get is a "burn in" screen (yellow fading into white) and the laptop fully crashes.
Any ideas for this one? Or do you think it's worth a bug report?

From your earlier comment, you said that you had the same card as Akuma (HD3650). Looking at http://wiki.gentoo.org/wiki/Radeon#Firmware, that means you'd need r600 ucode and not PITCAIRN which is for Southern Island cards. But it seems that there is no R600_smc.bin uploaded at http://people.freedesktop.org/~agd5f/radeon_ucode/. Some others at the Phoronix forums also ran across this problem. I guess that means that support for you card isn't implemented yet.

Last edited by Ledti (2013-06-28 23:52:01)

Akuma · 2013-06-29 02:28:42

Ledti wrote:

But it seems that there is no R600_smc.bin uploaded at http://people.freedesktop.org/~agd5f/radeon_ucode/. Some others at the Phoronix forums also ran across this problem. I guess that means that support for you card isn't implemented yet.

Yeah this makes sense. I thought the R600_rlc.bin was the only needed firmware but apparently a *_smc.bin is required for PM to function properly.
Well, I'll wait until this hits stable anyways. I'm happy enough to know that the days of Catalyst on my laptop are soon over .

Stunts · 2013-06-29 13:46:35

Ledti wrote:

From your earlier comment, you said that you had the same card as Akuma (HD3650). Looking at http://wiki.gentoo.org/wiki/Radeon#Firmware, that means you'd need r600 ucode and not PITCAIRN which is for Southern Island cards. But it seems that there is no R600_smc.bin uploaded at http://people.freedesktop.org/~agd5f/radeon_ucode/. Some others at the Phoronix forums also ran across this problem. I guess that means that support for you card isn't implemented yet.

Yeah, I copied all the firmware files under http://people.freedesktop.org/~agd5f/radeon_ucode/ to /usr/lib/firmware/radeon.
I was not sure whether my card had a code name or not. I guess it's just plain r600, which means no firmware.
I just thought it was odd to have such a spectacular (or not) failure. Guess I'll just have to wait and see...
Thank you anyway for your feedback, though.

Ledti · 2013-07-02 02:21:52

The latest upgrade to linux-agd5f-3.10.0-5 from 3.10.0-4 seems to have broken dpm for me. On my 6870, after the update, all I see are artifacts galore. I had to revert back to 3.10.0-4 to fix the problem.

It's kind of weird though because according to this user with the same card, the latest patches are working fine for him. I'm hesitant to report the regression in that thread because of this. Anyone else using linux-agd5f have the same problem with the latest patch set?

skottish · 2013-07-02 04:27:34

Ledti wrote:

It's kind of weird though because according to this user with the same card, the latest patches are working fine for him. I'm hesitant to report the regression in that thread because of this. Anyone else using linux-agd5f have the same problem with the latest patch set?

I can't help you with your problem directly, but since the AMD devs are using that thread at Phoronix to advance the open source drivers, I highly recommend that you post there. There's a very real chance that there may be open source drivers that compete with nVidia's drivers, so anything that can be helpful to them needs to be expressed. I understand that those forums can get ugly, but this is a major opportunity to be helpful.

Ledti · 2013-07-02 06:31:54

skottish wrote:

Ledti wrote:
It's kind of weird though because according to this user with the same card, the latest patches are working fine for him. I'm hesitant to report the regression in that thread because of this. Anyone else using linux-agd5f have the same problem with the latest patch set?
I can't help you with your problem directly, but since the AMD devs are using that thread at Phoronix to advance the open source drivers, I highly recommend that you post there. There's a very real chance that there may be open source drivers that compete with nVidia's drivers, so anything that can be helpful to them needs to be expressed. I understand that those forums can get ugly, but this is a major opportunity to be helpful.

Oh, I agree entirely. I was just waiting to see if the problem was an issue with the package rather than upstream. But I tested the two patchsets locally by compiling 3.10 twice, and the issue was still there. I've since posted to that thread relating as much.

skottish · 2013-07-03 05:07:00

Ledti wrote:

skottish wrote:
Ledti wrote:
It's kind of weird though because according to this user with the same card, the latest patches are working fine for him. I'm hesitant to report the regression in that thread because of this. Anyone else using linux-agd5f have the same problem with the latest patch set?
I can't help you with your problem directly, but since the AMD devs are using that thread at Phoronix to advance the open source drivers, I highly recommend that you post there. There's a very real chance that there may be open source drivers that compete with nVidia's drivers, so anything that can be helpful to them needs to be expressed. I understand that those forums can get ugly, but this is a major opportunity to be helpful.
Oh, I agree entirely. I was just waiting to see if the problem was an issue with the package rather than upstream. But I tested the two patchsets locally by compiling 3.10 twice, and the issue was still there. I've since posted to that thread relating as much.

Thanks for doing that. I know that your post didn't get a response yet, but it's a good thing. The fact that AMD may bring us open-source workstation class drivers is huge. Their harware is great. I've used nVidia cards for ten years under Linux, but I'd gladly switch if they can do this.

Arch Linux

#2451 2013-05-23 09:32:17

Re: Discussion: the ati/radeon open source drivers & radeon repository

#2452 2013-05-23 10:57:37

Re: Discussion: the ati/radeon open source drivers & radeon repository

#2453 2013-05-23 11:29:47

Re: Discussion: the ati/radeon open source drivers & radeon repository

#2454 2013-05-23 16:19:28

Re: Discussion: the ati/radeon open source drivers & radeon repository

#2455 2013-05-23 16:34:50

Re: Discussion: the ati/radeon open source drivers & radeon repository

#2456 2013-05-24 17:20:16

Re: Discussion: the ati/radeon open source drivers & radeon repository

#2457 2013-06-26 17:38:39

Re: Discussion: the ati/radeon open source drivers & radeon repository

#2458 2013-06-26 19:29:02

Re: Discussion: the ati/radeon open source drivers & radeon repository

#2459 2013-06-26 19:39:06

Re: Discussion: the ati/radeon open source drivers & radeon repository

#2460 2013-06-26 19:42:19

Re: Discussion: the ati/radeon open source drivers & radeon repository

#2461 2013-06-27 01:46:19

Re: Discussion: the ati/radeon open source drivers & radeon repository

#2462 2013-06-27 06:25:07

Re: Discussion: the ati/radeon open source drivers & radeon repository

#2463 2013-06-27 09:46:32

Re: Discussion: the ati/radeon open source drivers & radeon repository

#2464 2013-06-27 17:08:44

Re: Discussion: the ati/radeon open source drivers & radeon repository

#2465 2013-06-27 20:27:46

Re: Discussion: the ati/radeon open source drivers & radeon repository

#2466 2013-06-27 21:35:47

Re: Discussion: the ati/radeon open source drivers & radeon repository

#2467 2013-06-28 21:06:56

Re: Discussion: the ati/radeon open source drivers & radeon repository

#2468 2013-06-28 22:49:35

Re: Discussion: the ati/radeon open source drivers & radeon repository

#2469 2013-06-28 23:46:43

Re: Discussion: the ati/radeon open source drivers & radeon repository

#2470 2013-06-29 02:28:42

Re: Discussion: the ati/radeon open source drivers & radeon repository

#2471 2013-06-29 13:46:35

Re: Discussion: the ati/radeon open source drivers & radeon repository

#2472 2013-07-02 02:21:52

Re: Discussion: the ati/radeon open source drivers & radeon repository

#2473 2013-07-02 04:27:34

Re: Discussion: the ati/radeon open source drivers & radeon repository

#2474 2013-07-02 06:31:54

Re: Discussion: the ati/radeon open source drivers & radeon repository

#2475 2013-07-03 05:07:00

Re: Discussion: the ati/radeon open source drivers & radeon repository

Board footer