AMD radeon rx 550 Working?

[Christian]

the rx 550 works (3D!!) with android x86 7.1 r1 or higher???

[Mauro Rossi]

For 7.1 images they are available, amdgpu should work via gralloc_drm_pipe,

but I do not recall people having tested rx 550

Experimental oreo-x86 with kernel 4.14 and mesa-dev are here:

https://drive.google.com/drive/folders/0B_OFHiIqgpSFTFpkQWc1eXV3ME0

Please provide feedbacks (even just to know if they boot)

Mauro

[Christian]

HOW to Install this?? 

[Christian]

I will now test the 64 bit ISO and give some feedback, I have bought the Radeon RX 560.

Am Donnerstag, 30. November 2017 11:10:28 UTC+1 schrieb Christian:

[Christian]

Works ready nice! 

Am Donnerstag, 30. November 2017 11:10:28 UTC+1 schrieb Christian:

[Mauro Rossi]

Hi Christian,

I do not remember if those images are booting after installation,

however now I built a nougat-x86 image with amd dc,

it should work from GCN 2nd generation to Vega with gralloc_drm_pipe

I've uploaded on [1] an experimental nougat-x86 ISO which features:

- kernel 4.15rc6 with amd dc enabled for pre-Vega too

- latest AMD firmwares

- amdgpu driver enforced over radeon(si) CIK, VI parts

- llvm50

- latest mesa 17.4.0devel

- libdrm 2.4.89

...and..

- drm_hwcomposer with latest drm mode connector bugfixes (only in Live HWCOMP mode) - It is not stable, but needs testing too,

unfortunately my HD7950 GCN 1.0 is not supported by amd dc

I could anyway check drm_gralloc (Live CD DEBUG 'normal' drm_gralloc mode) and this time amdgpu worked like a charm with HD7950

So as a proposal, when you have time, I'd like to get feedback and dmesg/logcat for two test cases.

Test 1 - drm_gralloc -> boot with  "Live CD - Debug mode"

(This one is expected to boot from CIK to Vega)

Test 2 - drm_hwcomposer/gbm_gralloc  -> boot with "Live CD - Debug HWCOMP..."

(This one is booting on nouveau !!! Unbelivable, Intel does bootloop/cursor on top-left, amd dc is the only other driver with necessary capabilities, i.e atomic modesetting)

[1] https://drive.google.com/drive/folders/0B_OFHiIqgpSFMlk4bkhaMnAwbDA

file: android_x86_64_amd_dc_llvm50_mesa-17.4.0devel_hwc.iso

Mauro

[Christian]

Am Donnerstag, 30. November 2017 11:10:28 UTC+1 schrieb Christian:

[Mauro Rossi]

This one is oreo-x86 image with mesa 17.3.1, which is still not using AMD DC new driver and is based on drm_gralloc

I've uploaded and oreo-x86 image with mesa 17.4.0devel with kernel 4.15rc6 and AMD DC driver and with option of drm_hwcomposer.

https://drive.google.com/drive/folders/0B_OFHiIqgpSFTFpkQWc1eXV3ME0

Please, could someone with CIK, VI, Polaris or Vega GPU provide me feedback on the status of AMD DC driver,

in both use cases:

1) The usual "Live Debug mode" to see how AMD DC behaves with drm_gralloc

2) The experimental "Live Debug HWCOMP mode" to see how AMD DC behaves with drm_hwcomposer/gbm_gralloc

Thanks in advance

Mauro

[Christian]

Your image starts only until "Xpad ..." then I would have to enter something ....

Am Donnerstag, 30. November 2017 11:10:28 UTC+1 schrieb Christian:

[Mauro Rossi]

Il giorno sabato 6 gennaio 2018 19:39:15 UTC+1, Christian ha scritto:

Your image starts only until "Xpad ..." then I would have to enter something ....

In Live Debug modes: 

exit [enter]

exit [enter]

[Christian]

It is aborted with pixel image

Am Donnerstag, 30. November 2017 11:10:28 UTC+1 schrieb Christian:

[Mauro Rossi]

Il giorno sabato 6 gennaio 2018 20:05:50 UTC+1, Christian ha scritto:

It is aborted with pixel image

Thanks for feedback 

With both "Live Debug HWCOMP" and "Live Debug" boot attempts of oreo-x86 image?

https://drive.google.com/drive/folders/0B_OFHiIqgpSFTFpkQWc1eXV3ME0

What about "Live Debug HWCOMP" and "Live Debug" boot attempts of nougat-x86 image? (this one should also install in the more stable drm_gralloc mode)

https://drive.google.com/drive/folders/0B_OFHiIqgpSFMlk4bkhaMnAwbDA

Is there some Vega owner?

Mauro

[Christian]

oreo-x86 image with mesa 17.4.0devel with kernel 4.15rc6 Stop here

android_x86_64_amd_dc_llvm50_mesa-17.4.0devel_hwc.iso

Stops here too, but Android starts after a few seconds.

Am Donnerstag, 30. November 2017 11:10:28 UTC+1 schrieb Christian:

[youling 257]

I use my 4.14 kernel to boot oreo 8.1，failed，

request module fs-cpuset succeed，but still no fs ? traps: no user code available. invalid opcode:0000

在 2018年1月7日星期日 UTC+8上午4:36:08，Mauro Rossi写道：

[youling 257]

4.9 kernel，4.14 kernel

在 2018年1月7日星期日 UTC+8上午4:36:08，Mauro Rossi写道：

[Mauro Rossi]

Hi youling257,

in order to avoid the kernel crash you need to enable cpuset
as done in the following commit:

https://github.com/maurossi/linux/commit/72d89de1ae5fc6a0d526094acb6b9e89e77a8d2a

Mauro

[Christian]

Thanks, but I'm just a normal user, not a developer.

Am Donnerstag, 30. November 2017 11:10:28 UTC+1 schrieb Christian:

[yuki haya]

Hello.

I am using this image.

"Nougat_x86_64_amd_dc_llvm60_mesa - 18.1.0devel_vulkan.radv.iso"

How do I pass parameters to the kernel module?

I would like to enable AMDGPU DC with RX 560.

I specified "amdgpu.dc = 1" in GRUB.

/ sys / modules / amdgpu / parameters / "dc" is "-1 (Auto)".

I also tried "/ etc / modprobe.d /".

The content is "options amdgpu dc = 1"

Is this a known problem?

Thank you.

[Mauro Rossi]

Il giorno giovedì 29 marzo 2018 03:44:01 UTC+2, yuki haya ha scritto:

Hello.

I am using this image.

"Nougat_x86_64_amd_dc_llvm60_mesa - 18.1.0devel_vulkan.radv.iso"

How do I pass parameters to the kernel module?

I would like to enable AMDGPU DC with RX 560.

AMD_DC is already enabled in kernel config 

and also amdgpudrmfb is enforced over radeondrmfb,

were you able to boot?

radv is currently not working

[yuki haya]

Thank you.

I was able to start.

However, in the OpenGL application, the color of the texture is abnormal.

I will white out.

[Mauro Rossi]

Hi, I'm sending request for help to debug the current drm_hwcomposer + gbm_gralloc (HWCOMP=1 boot cmdline needed) with AMD GCN 2nd Gen Bonaire or higher

https://drive.google.com/drive/folders/0B_OFHiIqgpSFTFpkQWc1eXV3ME0

image: oreo_x86_k416_amd_dc_mesa-18.1.0_hwc1_revert_binder.iso

The main problem is that modesetting is not switching to graphical mode, console mode appears instead.

It would be great if someone could provide logcat and dmesg with following grub cmdline options:

HWCOMP=1 drm.debug=0x1e log_buf_len=1M

This will provide a mean to verify drm atomic ioctl messages and other drm issues as suggested in [1]

Please provide confirmation that you've seen drm debug content in dmesg output.

Mauro

[1] https://01.org/linuxgraphics/documentation/how-report-bugs

[yuki haya]

Hello.

I would like to help develop.

I got a Log in the way presented.

We will send Log at Polaris 11 (RX 580).

Will it be helpful?

Even in my environment it only started in console mode.

 

Thank you.

 

[Mauro Rossi]

Hi Yuki,

thanks a lot, I also tried to add drm.debug=0x1e grub cmdline,

but I do not see the expected output in dmesg log,

I was trying this to trouble shoot the problem in atomic mode setting.

Could you also send me /data/log.txt to check if hwc is also failing to set an atomic property in test mode?

Thanks a lot

Mauro

[yuki haya]

Hello. Mauro.

I checked / data / log.txt.

The atomic property does not seem to be set.

I will attach log.txt.

Thank you

Yuki

[Mauro Rossi]

Many thanks for your help

I see that amdgpudrmfb  was not initialized in this log.txt, 

there is another problem going on because amdgpudrmfb has been disabled,

while in the dmesg output amdgpudrmfb was active.

06-02 13:02:49.722     0     0 E         : [drm:amdgpu_init [amdgpu]] *ERROR* VGACON disables amdgpu kernel modesetting.

06-02 13:02:50.004     0     0 I uvesafb : fb0: VESA VGA frame buffer device

and then software renderer started:

06-02 13:02:50.782  2435  2435 I SurfaceFlinger: OpenGL ES informations: format=0x5

06-02 13:02:50.782  2435  2435 I SurfaceFlinger: vendor    : Google Inc.

06-02 13:02:50.782  2435  2435 I SurfaceFlinger: renderer  : Google SwiftShader

06-02 13:02:50.782  2435  2435 I SurfaceFlinger: version   : OpenGL ES 2.0 SwiftShader 4.0.0.3

Which grub menu option have you used? 

LiveCD DEBUG with addition of HWCOMP=1 variable should keep amdgpudrmfb active

Mauro

[Mauro Rossi]

I also saw now in the dmesg that firmwares are not loaded, 

but I checked and they are present in device/generic/firmware project

[    4.496481] amdgpu 0000:01:00.0: Direct firmware load for amdgpu/polaris10_pfp_2.bin failed with error -2

[    4.497131] amdgpu 0000:01:00.0: Direct firmware load for amdgpu/polaris10_me_2.bin failed with error -2

[    4.497275] amdgpu 0000:01:00.0: Direct firmware load for amdgpu/polaris10_ce_2.bin failed with error -2

[    4.498220] amdgpu 0000:01:00.0: Direct firmware load for amdgpu/polaris10_mec_2.bin failed with error -2

[    4.498819] amdgpu 0000:01:00.0: Direct firmware load for amdgpu/polaris10_mec2_2.bin failed with error -2

[yuki haya]

Thank you for your reply.

The grub option used normal startup mode.

I got a log in DEBUG mode earlier.

I will send it again.

Thank you.

[Mauro Rossi]

Hi Yuki,

I've seen your earlier post about amdgpu.dc=1 kernel module parameter

I think this may be the problem because I'm sure kernel modules parameters were processed before,

now they just seem to be ignored, also drm.debug=0x10 is just ignored

We may need help from Chih-Wei to solve this problem,

in the meantime I'm seeking the amdgpu.dc parameter in the kernel and HACK to enforce its default value to  "1 (Enabled)".

Mauro

[Mauro Rossi]

Hi,

Il giorno sabato 2 giugno 2018 20:36:42 UTC+2, Mauro Rossi ha scritto:

Hi Yuki,

I've seen your earlier post about amdgpu.dc=1 kernel module parameter

I think this may be the problem because I'm sure kernel modules parameters were processed before,

now they just seem to be ignored, also drm.debug=0x10 is just ignored

amdgpu.dc setting is not problematic because otherwise atomic would not be available at all

in order to set drm.debug=16 to get atomic messages is possible with following procedure:

Boot in LiveCD DEBUG mode (with HWCOMP=1 parameter)

exit

exit

[ALT]+[F1]

stop[ENTER]  (NOTE: it will not be echoed to screen)

echo 16 > /sys/module/drm/parameters/debug

start[ENTER]  (wait 30 seconds)

[ALT]+[F1]

stop[ENTER]  (NOTE: it will not be echoed to screen)

At this point dmesg and /data/log.txt will both have the drm atomic debug messages

Mine for HD7790 Bonaire are in the attachment, it seams that the problem is an unsupported format AB24  little-endian,

which causes the -EINVAL (-22) error.

Most probably you have the same error

Mauro

...

06-02 23:33:17.415     0     0 D         : [drm:drm_atomic_state_init [drm]] Allocated atomic state 28100ec5

06-02 23:33:17.415     0     0 D         : [drm:drm_atomic_get_plane_state [drm]] Added [PLANE:41:plane-5] b79a83bb state to 28100ec5

06-02 23:33:17.415     0     0 D         : [drm:drm_atomic_get_crtc_state [drm]] Added [CRTC:43:crtc-0] 8b04e189 state to 28100ec5

06-02 23:33:17.415     0     0 D         : [drm:drm_atomic_set_crtc_for_plane [drm]] Link plane state b79a83bb to [CRTC:43:crtc-0]

06-02 23:33:17.415     0     0 D         : [drm:drm_atomic_set_fb_for_plane [drm]] Set [FB:78] for plane state b79a83bb

06-02 23:33:17.415     0     0 D         : [drm:drm_atomic_set_mode_prop_for_crtc [drm]] Set [MODE:1024x768] for CRTC state 8b04e189

06-02 23:33:17.415     0     0 D         : [drm:drm_atomic_get_connector_state [drm]] Added [CONNECTOR:61:DVI-D-2] f7676eb8 state to 28100ec5

06-02 23:33:17.415     0     0 D         : [drm:drm_atomic_check_only [drm]] checking 28100ec5

06-02 23:33:17.415     0     0 D         : [drm:drm_atomic_check_only [drm]] Invalid pixel format AB24 little-endian (0x34324241), modifier 0x0

06-02 23:33:17.415  4922  4922 I hwc-drm-display-compositor: Commit test pset failed ret=-22

06-02 23:33:17.415  4922  4922 I hwc-drm-display-compositor: Commit test failed, squashing frame for display 0

06-02 23:33:17.415  4922  4922 E hwc-drm-display-compositor: Composite failed for display 0

06-02 23:33:17.415  4922  4922 E hwc-drm-two: Failed to apply the frame composition ret=-22

06-02 23:33:17.415  4922  4922 E HWComposer: presentAndGetReleaseFences: failed for display 0: BadParameter (4)

06-02 23:33:17.415     0     0 D [drm:drm_atomic_check_only [drm]] [PLANE: 41:plane-5] atomic core check failed

06-02 23:33:17.415     0     0 D         : [drm:drm_atomic_state_default_clear [drm]] Clearing atomic state 28100ec5

06-02 23:33:17.415     0     0 D         : [drm:__drm_atomic_state_free [drm]] Freeing atomic state 28100ec5

...

[yuki haya]

Hello.

Logged by Mr. Mauro 's method.

The same problem also occurs in RX 580.

I will attach a log.

I wish for problem solving.

Thank you.

[Mauro Rossi]

The problem is that the hwcomposer related code in surfaceflinger 

supports only HAL_PIXEL_FORMAT_RGBA_8888

The final solution is either having the driver support RGBA 

or surfaceflinger support selection or fallback to HAL_PIXEL_FORMAT_BGRA_8888 in hwcomposer path.

In frameworks/native RenderEngine and eglApi there are similar capabilities: [1] and [2]

As a first prototypical attempt I have replaced the hardcoded HAL pixel format with HAL_PIXEL_FORMAT_BGRA_8888 [3]

The first results are promising as I could finally boot with Bonaire HD7790 on the drm_hwcomposer+gbm_gralloc stack

and by correcting a problem in mesa radv, the vulkan API is enabled.

Far from perfect, because there are still some issues:

- (tiling?) artifacts, visible in attachment, in V1 Vulkan API benchmark, PPSSPP, SaschaWillems Vulkan demos (Gears, Texture)

- crashes SIGSEGV MAPERR with 3Dbench vulkan (Slingshot Eextreme)

- problems with houdini/native bridge for ARM apps SaschaWillems VulkanCapsViewer,  ES2 Basemark

- youtube videos show black surfaces

but on the other hand the OpenGL GLES 3.x apps are not crashing, Antutu 3D can complete the benchmark

64bit Images being uploaded here: https://drive.google.com/drive/folders/0B_OFHiIqgpSFTFpkQWc1eXV3ME0

hwc1, kernel 4.16, llvm70, mesa 18.1.1: remember HWCOMP=1 VULKAN=radv in grub cmdline

hwc2 robh, kernel 4.17, llvm70, mesa-dev: remember HWCOMP=2robh VULKAN=radv in grub cmdline

Mauro

[1] http://git.osdn.net/view?p=android-x86/frameworks-native.git;a=commit;h=792c8dc009bd3a0c44eb39e757a95e099c03b54c

[2] http://git.osdn.net/view?p=android-x86/frameworks-native.git;a=commit;h=519828c5b649e5e83f18444666f0672ab7852518

[3] https://github.com/maurossi/frameworks_native/commits/BGRA_in_hwc2

[Mauro Rossi]

Hi there,

The problem is solved by implementing the missing GRALLOC_USAGE_HW_COMPOSER usage flag mappings in gbm_gralloc,

with the attached patch.

At the following link an iso

nougat-x86 64bit with hwc1, kernel 4.18rc2, llvm70, mesa 18.1.3 updated amdgpu firmwares and init_hal_vulkan() taking care of loading the correct vulkan hal

https://drive.google.com/drive/folders/0B_OFHiIqgpSFMlk4bkhaMnAwbDA

Mauro

[Toetje583(NL)]

I have a R7 260X (Bonaire) and a R9 270, both cards only boot to console, anything I can assist you with?

Best Regards,

Toetje

Op zaterdag 30 juni 2018 13:22:09 UTC+2 schreef Mauro Rossi:

[Mauro Rossi]

Hi,

Il giorno domenica 1 luglio 2018 22:48:54 UTC+2, Toetje583(NL) ha scritto:

I have a R7 260X (Bonaire) and a R9 270, both cards only boot to console, anything I can assist you with?

Best Regards,

Toetje

Most probably the kernels 4.17 and 4.18rc2 are freezing 

I've uploaded oreo-x86 images with kernel 4.16, llvm70, mesa-dev (which dropped flink names - i.e gralloc.drm  support - but enabled dma-bufs/prime fd and hwcomposer.drm + gralloc.gbm for intel)

Having hwcomposer.drm + gralloc.gbm working with i965 gifted us vulkan anv (impressively stable) together with amdgpu (amd dc) & vulkan radv

Android O seem more "vulkan ready" than Android N 

https://drive.google.com/drive/folders/0B_OFHiIqgpSFTFpkQWc1eXV3ME0

NOTES:

R9 270 is GCN 1st gen not supporting amd dc, so it will not boot because atomic is not available

R7 260X (Bonaire) will work as the HD7790 I used for testing

nouveau works with hwcomposer.drm + gralloc.gbm

radeon and svga (vmwgfx) do not work, surfaceflinger cannot open framebuffer with dma-bufs/prime fd

07-01 18:41:13.941  2739  2739 E SurfaceFlinger: hwcomposer module not found

07-01 18:41:13.941  2739  2739 E SurfaceFlinger: ERROR: failed to open framebuffer (Invalid argument), aborting

--------- beginning of crash

07-01 18:41:13.941  2739  2739 F libc    : Fatal signal 6 (SIGABRT), code -6 in tid 2739 (surfaceflinger)

Solving this last problem will achieve a great result: oreo-x86 working with gralloc.gbm and dma-bufs,

in order to avoid regression in radeon and vmwgfx

Mauro

[Toetje583(NL)]

RX 260X,

Used: oreo_x86_64_8.1-rc1_k416_llvm70_mesa-18.2.0-devel_w26_prime_fd_BGRA_vulkan.iso

Its loading just the GUI isent getting started it seems, here a dmesg:

https://pastebin.com/8J582NWM

Best Regards,

Op maandag 2 juli 2018 01:36:34 UTC+2 schreef Mauro Rossi:

[Toetje583(NL)]

Ah, the live image was just not working, I created a new partition and added it to grub, it´s starting now but i´m experiencing lots of graphical glitches!

Included the log.txt :)


Op maandag 2 juli 2018 08:43:25 UTC+2 schreef Toetje583(NL):

[Toetje583(NL)]

I dump my logs here:

https://drive.google.com/open?id=1BbtwfgNA9ZgwcU46pvzPvRpBKjMB74_0

Also tested Intel HD 530, boots fine just lots of apps not starting.

Op maandag 2 juli 2018 09:13:13 UTC+2 schreef Toetje583(NL):

[Mauro Rossi]

Hi,

please try to set to vmalloc=192M or better vmalloc=256M

and collect also output of 'dumpsys SurfaceFlinger'

Instead of using drm.debug=16 which does not work with android-x86,

to get drm debug messages use the following procedure on R7 260X

Boot in DEBUG mode

modprobe amdgpu

echo 30 > /sys/module/drm/parameters/debug

exit

exit

[Toetje583(NL)]

Seems like  vmalloc=256M did the trick, no more crashes or artifacts, was even able to complete a Antutu Benchmark score: 526721

CPU: 200914
GPU: 198282
UX: 107159
MEM: 20366

New logs:

https://drive.google.com/open?id=1BbtwfgNA9ZgwcU46pvzPvRpBKjMB74_0


Op maandag 2 juli 2018 10:03:41 UTC+2 schreef Mauro Rossi:

[Toetje583(NL)]

Forgot to add, seems like the first Antutu benchmark failed, this is the Refinery benchmark ES3.1+AEP, however coastline worked also ES3.1+AEP.

Op maandag 2 juli 2018 11:22:57 UTC+2 schreef Toetje583(NL):

[yuki haya]

Hello.

Thank you for the new ISO image.

We will inform you of results in 4790k + B85 + RX580 environment.

In debug mode the following was applied.

"HWCOMP = 1", "VULKAN = radv" and "vmalloc = 256 M"

I will attach a log.

Antutu Benchmark is completed.

Total Score: 909459

CPU: 185725

GPU: 578840

UX: 135450

MEM: 9444

Thank you.

= Yuki =

[Mauro Rossi]

Just FYI, there are changes that I forgot to mention before, thanks to Yuki for pointing with a question.

With these new oreo-x86 images hwcomposer.drm + gralloc.gbm have been enabled for intel, amdgpu and nouveau [1]

so HWCOMP=1 boot cmdline is not necessary anymore,

vulkan hal is selected in init.sh based on the loaded driver, so VULKAN=radv is not used

If HWCOMP or VULKAN are defined they have no effect in these oreo-x86 isos

Not all Intel chipsets are supported by vulkan anv Intel HD 530 may be one of those,but I'm not sure,

I tested with Lenovo T460 i965 Skylake GT2

Feedbacks on vulkan radv from people with GCN 2nd Gen and later are appreciated especially for:

Vega Gen 5

Ryzen APU

Kaveri APU

(32bit) oreo_x86_8.1-rc1_k416_llvm70_mesa-18.2.0-devel_w26_prime_fd_BGRA_vulkan

(64bit) oreo_x86_64_8.1-rc1_k416_llvm70_mesa-18.2.0-devel_w26_prime_fd_BGRA_vulkan

The corresponding nouveau-x86 images with 4.16 kernel have more problems than oreo-x86 ones,

since oreo-x86 can now be installed on Sata HDD I will keep up testing on oreo-x86

Having surfaceflinger to able to deal directly with gralloc.gbm (or vice versa) would pave the way for oreo-x86 release 1 with hwcomposer.drm + gralloc.gbm for all enabled drivers

This milestone was desired from quite some time, at least from me :-) 

Now some results with oreo-x86_64_8.1rc1_*_prime_fd_BGRA_vulkan with both anv and radv:

[working apps]

V1 Benchmark (Vulkan API) - should work

3D Bench (Slinghshot extreeme with Vulkan API) - should work

PPSSPP - should work

Toy Zombies: Lite - should work (I cannot understand the commands with mouse cursor I always get killed)

Olympus Rising - should work (very nice)

https://github.com/googlesamples/android-vulkan-tutorials - working

VulkanCapsViewer - use 1.6 apk for x86 from http://vulkan.gpuinfo.org/download.php

Most of SaschaWillems Vulkan examples http://vulkan.gpuinfo.org/downloads/examples/vulkan_examples_android.7z

[crashing apps - in anv and maybe also in radv - but to be verified]

Following SaschaWillems Vulkan examples 

adb connect [IP address]

adb install -r ~/Downloads/vulkan_examples_android/vulkanDeferredmulitsampling.apk

adb install -r ~/Downloads/vulkan_examples_android/vulkanIndirectdraw.apk

adb install -r ~/Downloads/vulkan_examples_android/vulkanInstancing.apk

adb install -r ~/Downloads/vulkan_examples_android/vulkanTerraintessellation.apk

Mauro

[Toetje583(NL)]

I will test them aswell and report back, if you need 20 core 40 treats server for compiling etc let me know, i have one machine that is doing nothing atm.
If you need a person to test stuff regarding Bonaire GPU, or even intel and NVIDIA I can assist you :)

I have,

GTX 1030
GTX 1060
GTX 1070
HD7870
R9 270
R7 260X
Intel HD 530
AMD E-350 APU with Radeon HD6310 (Kinda OLD)

Server specs:

2X Xeon E5 2680v2 10 core each 20 total 40 treaths

32GB Memory

2X 120GB SSD
1X 3TB Drive
1X 500GB Drive


Best Regards,




Op maandag 2 juli 2018 13:58:44 UTC+2 schreef Mauro Rossi:

[Toetje583(NL)]

R7 260X

V1 Benchmark (Vulkan API) - (Works but textures are bouncing)

3D Bench (Slinghshot extreeme with Vulkan API) - (Works but textures are bouncing)

PPSSPP - Wanted to test but need to get some isos first, the one i had dident work but also dident work with Gles.

Toy Zombies: Lite - Works (Died fast xD)

Olympus Rising - Works
VulkanCapsViewer - Works


Apps crashing:

King of Avalon: Dragon Warfare

Op maandag 2 juli 2018 13:58:44 UTC+2 schreef Mauro Rossi:

Just FYI, there are changes that I forgot to mention before, thanks to Yuki for pointing with a question.

[Mauro Rossi]

Hi, 

Il giorno lunedì 2 luglio 2018 19:21:07 UTC+2, Toetje583(NL) ha scritto:

R7 260X

V1 Benchmark (Vulkan API) - (Works but textures are bouncing)

3D Bench (Slinghshot extreeme with Vulkan API) - (Works but textures are bouncing)

Do you have screenshot or video, to see what is the problem? The tiling artifacts should be solved with gbm_gralloc patch [1]

[1] https://github.com/maurossi/gbm_gralloc/commit/6ea56d69f008d5aa9513650d8b66aef9fe51bd24

[Mauro Rossi]

Hi,

Il giorno lunedì 2 luglio 2018 17:51:24 UTC+2, Toetje583(NL) ha scritto:

I will test them aswell and report back, if you need 20 core 40 treats server for compiling etc let me know, i have one machine that is doing nothing atm.

Interesting could you provide me instructions on how to privately access via remote, 

in many cases I build with AMD Phenom(tm) II X3 720  or HP Prolian ML350 G5 (when lucky)

also uploading test iso from there may help

I will leave as soon as you need it back

 

 

If you need a person to test stuff regarding Bonaire GPU, or even intel and NVIDIA I can assist you :)

nouveau may require some Android CTS dEQP test run for EGL, GLES2 and GLES3/3.1

as we never tested on GTX 10x0

I don't know how stable it is with dma-bufs

Also I'm trying to see if gralloc.gbm or chromeos or intel minigbm gralloc0 may work for radeon and vmwgfx,

but I still need to dig in it

[Toetje583(NL)]

Do you use Discord or skype so we can discuss this discuss this further:)

Best Regards,



Op maandag 2 juli 2018 21:41:12 UTC+2 schreef Mauro Rossi:

[Chih-Wei Huang]

2018-07-02 19:58 GMT+08:00 Mauro Rossi <issor...@gmail.com>:
>
> (64bit)
> oreo_x86_64_8.1-rc1_k416_llvm70_mesa-18.2.0-devel_w26_prime_fd_BGRA_vulkan

Hi Mauro,
I made some tests with this iso on devices with different GPUs I have.
The result:

* Apollo Lake: boot and run fine.
* SkyLake: bootanimation appear but the screen is messed up
* Coffee Lake: Reboot immediately on switching to graphic mode
* Cherrytrail: Only black screen on switching to graphic mode
* Nvidia GTX1060: Had to set atomic=1 on loading nouveau driver,
then bootanimation ok, after boot complete, no mouse cursor
and most icons disappear
* QEMU virgl: bootanimation ok, but most parts of the screen
are black after boot complete

It's just quick tests to see what work or not.
I haven't checked any log yet.

I'm a little disappointed about the results of Intel GPUs (i965)
since they are so variant between different GPUs...

[Mauro Rossi]

Hi Chih-Wei, 

Il giorno venerdì 13 luglio 2018 11:47:35 UTC+2, Chih-Wei Huang ha scritto:

2018-07-02 19:58 GMT+08:00 Mauro Rossi <issor...@gmail.com>:
>
> (64bit)
> oreo_x86_64_8.1-rc1_k416_llvm70_mesa-18.2.0-devel_w26_prime_fd_BGRA_vulkan

Hi Mauro,
I made some tests with this iso on devices with different GPUs I have.
The result:

* Apollo Lake: boot and run fine.
* SkyLake: bootanimation appear but the screen is messed up

What Skylake model? On Intel 520 (GT2) it works

 

* Coffee Lake: Reboot immediately on switching to graphic mode
* Cherrytrail: Only black screen on switching to graphic mode

On Cherrytrail ( Chuwi Hi12 ) I see that it does not support BGRA (AR24) but I do not know what is supported, because also RGBA (AB24) is not supported either in atomic

I've started to build modetest, proptest, vbltest and modeprint in libdrm to have some mean to debug the formats supported

 

* Nvidia GTX1060: Had to set atomic=1 on loading nouveau driver,
   then bootanimation ok, after boot complete, no mouse cursor
   and most icons disappear

Strange, it was hardcoded in kernel

and on GT610 it boots, even if vending crashes at any serch in vending app

 

* QEMU virgl: bootanimation ok, but most parts of the screen
   are black after boot complete

I've done an attempt to enable hwcomposer.drm + gralloc.gbm for vmwgfx and I boots, no cursor or invisible cursor (I suspect the gralloc.gbm GRALLOC_USAGE_CURSOR commented may be 

 

It's just quick tests to see what work or not.
I haven't checked any log yet.

I'm a little disappointed about the results of Intel GPUs (i965)
since they are so variant between different GPUs...

I've also seen that Intel HD 3000 support of pixel format is very different from Skylake GT2  and Cherrytrail pixel formats,

we need to find a way to probe and fallback in surfaceflinger

Mauro

[Mauro Rossi]

Hi,

I've uploaded a new set of images featuring RGBA_8888 in amd dc.

Results are promising, but we need some help in the forum to check hwc1 and hwc2 images (without enforced BGRA)

on RX550, RX580, Ryzen and other AMD DC supported GPU/APU

Feedbacks on Vulkan enabled apps are appreciated

https://drive.google.com/drive/folders/0B_OFHiIqgpSFTFpkQWc1eXV3ME0

hwc2 images to test are the ones without HWCOMP=2,2robh in the name, 

these do not require parameters , except vmalloc=256M for big screens ( >= 1920 x 1080)

As a side note tests on vmwgfx which has been enabled in these images [1] and [2]

show results similar to the ones Chih-Wei had  have with virgl (black screen with visible status bar and navigation bar)

Mauro

[1] https://github.com/maurossi/device_generic_common/commits/oreo-x86_vulkan

[2] https://github.com/maurossi/device_generic_common/commits/oreo-x86_vulkan_hwc2